fs/ceph/inode.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/ceph/ceph_debug.h>
   3
   4 #include <linux/module.h>
   5 #include <linux/fs.h>
   6 #include <linux/slab.h>
   7 #include <linux/string.h>
   8 #include <linux/uaccess.h>
   9 #include <linux/kernel.h>
  10 #include <linux/writeback.h>
  11 #include <linux/vmalloc.h>
  12 #include <linux/xattr.h>
  13 #include <linux/posix_acl.h>
  14 #include <linux/random.h>
  15 #include <linux/sort.h>
  16 #include <linux/iversion.h>
  17
  18 #include "super.h"
  19 #include "mds_client.h"
  20 #include "cache.h"
  21 #include <linux/ceph/decode.h>
  22
  23 /*
  24  * Ceph inode operations
  25  *
  26  * Implement basic inode helpers (get, alloc) and inode ops (getattr,
  27  * setattr, etc.), xattr helpers, and helpers for assimilating
  28  * metadata returned by the MDS into our cache.
  29  *
  30  * Also define helpers for doing asynchronous writeback, invalidation,
  31  * and truncation for the benefit of those who can't afford to block
  32  * (typically because they are in the message handler path).
  33  */
  34
  35 static const struct inode_operations ceph_symlink_iops;
  36
  37 static void ceph_inode_work(struct work_struct *work);
  38
  39 /*
  40  * find or create an inode, given the ceph ino number
  41  */
  42 static int ceph_set_ino_cb(struct inode *inode, void *data)
  43 {
  44         struct ceph_inode_info *ci = ceph_inode(inode);
  45         struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
  46
  47         ci->i_vino = *(struct ceph_vino *)data;
  48         inode->i_ino = ceph_vino_to_ino_t(ci->i_vino);
  49         inode_set_iversion_raw(inode, 0);
  50         percpu_counter_inc(&mdsc->metric.total_inodes);
  51
  52         return 0;
  53 }
  54
  55 struct inode *ceph_get_inode(struct super_block *sb, struct ceph_vino vino)
  56 {
  57         struct inode *inode;
  58
  59         if (ceph_vino_is_reserved(vino))
  60                 return ERR_PTR(-EREMOTEIO);
  61
  62         inode = iget5_locked(sb, (unsigned long)vino.ino, ceph_ino_compare,
  63                              ceph_set_ino_cb, &vino);
  64         if (!inode)
  65                 return ERR_PTR(-ENOMEM);
  66
  67         dout("get_inode on %llu=%llx.%llx got %p new %d\n", ceph_present_inode(inode),
  68              ceph_vinop(inode), inode, !!(inode->i_state & I_NEW));
  69         return inode;
  70 }
  71
  72 /*
  73  * get/constuct snapdir inode for a given directory
  74  */
  75 struct inode *ceph_get_snapdir(struct inode *parent)
  76 {
  77         struct ceph_vino vino = {
  78                 .ino = ceph_ino(parent),
  79                 .snap = CEPH_SNAPDIR,
  80         };
  81         struct inode *inode = ceph_get_inode(parent->i_sb, vino);
  82         struct ceph_inode_info *ci = ceph_inode(inode);
  83
  84         if (IS_ERR(inode))
  85                 return inode;
  86
  87         if (!S_ISDIR(parent->i_mode)) {
  88                 pr_warn_once("bad snapdir parent type (mode=0%o)\n",
  89                              parent->i_mode);
  90                 goto err;
  91         }
  92
  93         if (!(inode->i_state & I_NEW) && !S_ISDIR(inode->i_mode)) {
  94                 pr_warn_once("bad snapdir inode type (mode=0%o)\n",
  95                              inode->i_mode);
  96                 goto err;
  97         }
  98
  99         inode->i_mode = parent->i_mode;
 100         inode->i_uid = parent->i_uid;
 101         inode->i_gid = parent->i_gid;
 102         inode->i_mtime = parent->i_mtime;
 103         inode->i_ctime = parent->i_ctime;
 104         inode->i_atime = parent->i_atime;
 105         ci->i_rbytes = 0;
 106         ci->i_btime = ceph_inode(parent)->i_btime;
 107
 108         if (inode->i_state & I_NEW) {
 109                 inode->i_op = &ceph_snapdir_iops;
 110                 inode->i_fop = &ceph_snapdir_fops;
 111                 ci->i_snap_caps = CEPH_CAP_PIN; /* so we can open */
 112                 unlock_new_inode(inode);
 113         }
 114
 115         return inode;
 116 err:
 117         if ((inode->i_state & I_NEW))
 118                 discard_new_inode(inode);
 119         else
 120                 iput(inode);
 121         return ERR_PTR(-ENOTDIR);
 122 }
 123
 124 const struct inode_operations ceph_file_iops = {
 125         .permission = ceph_permission,
 126         .setattr = ceph_setattr,
 127         .getattr = ceph_getattr,
 128         .listxattr = ceph_listxattr,
 129         .get_acl = ceph_get_acl,
 130         .set_acl = ceph_set_acl,
 131 };
 132
 133
 134 /*
 135  * We use a 'frag tree' to keep track of the MDS's directory fragments
 136  * for a given inode (usually there is just a single fragment).  We
 137  * need to know when a child frag is delegated to a new MDS, or when
 138  * it is flagged as replicated, so we can direct our requests
 139  * accordingly.
 140  */
 141
 142 /*
 143  * find/create a frag in the tree
 144  */
 145 static struct ceph_inode_frag *__get_or_create_frag(struct ceph_inode_info *ci,
 146                                                     u32 f)
 147 {
 148         struct rb_node **p;
 149         struct rb_node *parent = NULL;
 150         struct ceph_inode_frag *frag;
 151         int c;
 152
 153         p = &ci->i_fragtree.rb_node;
 154         while (*p) {
 155                 parent = *p;
 156                 frag = rb_entry(parent, struct ceph_inode_frag, node);
 157                 c = ceph_frag_compare(f, frag->frag);
 158                 if (c < 0)
 159                         p = &(*p)->rb_left;
 160                 else if (c > 0)
 161                         p = &(*p)->rb_right;
 162                 else
 163                         return frag;
 164         }
 165
 166         frag = kmalloc(sizeof(*frag), GFP_NOFS);
 167         if (!frag)
 168                 return ERR_PTR(-ENOMEM);
 169
 170         frag->frag = f;
 171         frag->split_by = 0;
 172         frag->mds = -1;
 173         frag->ndist = 0;
 174
 175         rb_link_node(&frag->node, parent, p);
 176         rb_insert_color(&frag->node, &ci->i_fragtree);
 177
 178         dout("get_or_create_frag added %llx.%llx frag %x\n",
 179              ceph_vinop(&ci->netfs.inode), f);
 180         return frag;
 181 }
 182
 183 /*
 184  * find a specific frag @f
 185  */
 186 struct ceph_inode_frag *__ceph_find_frag(struct ceph_inode_info *ci, u32 f)
 187 {
 188         struct rb_node *n = ci->i_fragtree.rb_node;
 189
 190         while (n) {
 191                 struct ceph_inode_frag *frag =
 192                         rb_entry(n, struct ceph_inode_frag, node);
 193                 int c = ceph_frag_compare(f, frag->frag);
 194                 if (c < 0)
 195                         n = n->rb_left;
 196                 else if (c > 0)
 197                         n = n->rb_right;
 198                 else
 199                         return frag;
 200         }
 201         return NULL;
 202 }
 203
 204 /*
 205  * Choose frag containing the given value @v.  If @pfrag is
 206  * specified, copy the frag delegation info to the caller if
 207  * it is present.
 208  */
 209 static u32 __ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 210                               struct ceph_inode_frag *pfrag, int *found)
 211 {
 212         u32 t = ceph_frag_make(0, 0);
 213         struct ceph_inode_frag *frag;
 214         unsigned nway, i;
 215         u32 n;
 216
 217         if (found)
 218                 *found = 0;
 219
 220         while (1) {
 221                 WARN_ON(!ceph_frag_contains_value(t, v));
 222                 frag = __ceph_find_frag(ci, t);
 223                 if (!frag)
 224                         break; /* t is a leaf */
 225                 if (frag->split_by == 0) {
 226                         if (pfrag)
 227                                 memcpy(pfrag, frag, sizeof(*pfrag));
 228                         if (found)
 229                                 *found = 1;
 230                         break;
 231                 }
 232
 233                 /* choose child */
 234                 nway = 1 << frag->split_by;
 235                 dout("choose_frag(%x) %x splits by %d (%d ways)\n", v, t,
 236                      frag->split_by, nway);
 237                 for (i = 0; i < nway; i++) {
 238                         n = ceph_frag_make_child(t, frag->split_by, i);
 239                         if (ceph_frag_contains_value(n, v)) {
 240                                 t = n;
 241                                 break;
 242                         }
 243                 }
 244                 BUG_ON(i == nway);
 245         }
 246         dout("choose_frag(%x) = %x\n", v, t);
 247
 248         return t;
 249 }
 250
 251 u32 ceph_choose_frag(struct ceph_inode_info *ci, u32 v,
 252                      struct ceph_inode_frag *pfrag, int *found)
 253 {
 254         u32 ret;
 255         mutex_lock(&ci->i_fragtree_mutex);
 256         ret = __ceph_choose_frag(ci, v, pfrag, found);
 257         mutex_unlock(&ci->i_fragtree_mutex);
 258         return ret;
 259 }
 260
 261 /*
 262  * Process dirfrag (delegation) info from the mds.  Include leaf
 263  * fragment in tree ONLY if ndist > 0.  Otherwise, only
 264  * branches/splits are included in i_fragtree)
 265  */
 266 static int ceph_fill_dirfrag(struct inode *inode,
 267                              struct ceph_mds_reply_dirfrag *dirinfo)
 268 {
 269         struct ceph_inode_info *ci = ceph_inode(inode);
 270         struct ceph_inode_frag *frag;
 271         u32 id = le32_to_cpu(dirinfo->frag);
 272         int mds = le32_to_cpu(dirinfo->auth);
 273         int ndist = le32_to_cpu(dirinfo->ndist);
 274         int diri_auth = -1;
 275         int i;
 276         int err = 0;
 277
 278         spin_lock(&ci->i_ceph_lock);
 279         if (ci->i_auth_cap)
 280                 diri_auth = ci->i_auth_cap->mds;
 281         spin_unlock(&ci->i_ceph_lock);
 282
 283         if (mds == -1) /* CDIR_AUTH_PARENT */
 284                 mds = diri_auth;
 285
 286         mutex_lock(&ci->i_fragtree_mutex);
 287         if (ndist == 0 && mds == diri_auth) {
 288                 /* no delegation info needed. */
 289                 frag = __ceph_find_frag(ci, id);
 290                 if (!frag)
 291                         goto out;
 292                 if (frag->split_by == 0) {
 293                         /* tree leaf, remove */
 294                         dout("fill_dirfrag removed %llx.%llx frag %x"
 295                              " (no ref)\n", ceph_vinop(inode), id);
 296                         rb_erase(&frag->node, &ci->i_fragtree);
 297                         kfree(frag);
 298                 } else {
 299                         /* tree branch, keep and clear */
 300                         dout("fill_dirfrag cleared %llx.%llx frag %x"
 301                              " referral\n", ceph_vinop(inode), id);
 302                         frag->mds = -1;
 303                         frag->ndist = 0;
 304                 }
 305                 goto out;
 306         }
 307
 308
 309         /* find/add this frag to store mds delegation info */
 310         frag = __get_or_create_frag(ci, id);
 311         if (IS_ERR(frag)) {
 312                 /* this is not the end of the world; we can continue
 313                    with bad/inaccurate delegation info */
 314                 pr_err("fill_dirfrag ENOMEM on mds ref %llx.%llx fg %x\n",
 315                        ceph_vinop(inode), le32_to_cpu(dirinfo->frag));
 316                 err = -ENOMEM;
 317                 goto out;
 318         }
 319
 320         frag->mds = mds;
 321         frag->ndist = min_t(u32, ndist, CEPH_MAX_DIRFRAG_REP);
 322         for (i = 0; i < frag->ndist; i++)
 323                 frag->dist[i] = le32_to_cpu(dirinfo->dist[i]);
 324         dout("fill_dirfrag %llx.%llx frag %x ndist=%d\n",
 325              ceph_vinop(inode), frag->frag, frag->ndist);
 326
 327 out:
 328         mutex_unlock(&ci->i_fragtree_mutex);
 329         return err;
 330 }
 331
 332 static int frag_tree_split_cmp(const void *l, const void *r)
 333 {
 334         struct ceph_frag_tree_split *ls = (struct ceph_frag_tree_split*)l;
 335         struct ceph_frag_tree_split *rs = (struct ceph_frag_tree_split*)r;
 336         return ceph_frag_compare(le32_to_cpu(ls->frag),
 337                                  le32_to_cpu(rs->frag));
 338 }
 339
 340 static bool is_frag_child(u32 f, struct ceph_inode_frag *frag)
 341 {
 342         if (!frag)
 343                 return f == ceph_frag_make(0, 0);
 344         if (ceph_frag_bits(f) != ceph_frag_bits(frag->frag) + frag->split_by)
 345                 return false;
 346         return ceph_frag_contains_value(frag->frag, ceph_frag_value(f));
 347 }
 348
 349 static int ceph_fill_fragtree(struct inode *inode,
 350                               struct ceph_frag_tree_head *fragtree,
 351                               struct ceph_mds_reply_dirfrag *dirinfo)
 352 {
 353         struct ceph_inode_info *ci = ceph_inode(inode);
 354         struct ceph_inode_frag *frag, *prev_frag = NULL;
 355         struct rb_node *rb_node;
 356         unsigned i, split_by, nsplits;
 357         u32 id;
 358         bool update = false;
 359
 360         mutex_lock(&ci->i_fragtree_mutex);
 361         nsplits = le32_to_cpu(fragtree->nsplits);
 362         if (nsplits != ci->i_fragtree_nsplits) {
 363                 update = true;
 364         } else if (nsplits) {
 365                 i = prandom_u32_max(nsplits);
 366                 id = le32_to_cpu(fragtree->splits[i].frag);
 367                 if (!__ceph_find_frag(ci, id))
 368                         update = true;
 369         } else if (!RB_EMPTY_ROOT(&ci->i_fragtree)) {
 370                 rb_node = rb_first(&ci->i_fragtree);
 371                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 372                 if (frag->frag != ceph_frag_make(0, 0) || rb_next(rb_node))
 373                         update = true;
 374         }
 375         if (!update && dirinfo) {
 376                 id = le32_to_cpu(dirinfo->frag);
 377                 if (id != __ceph_choose_frag(ci, id, NULL, NULL))
 378                         update = true;
 379         }
 380         if (!update)
 381                 goto out_unlock;
 382
 383         if (nsplits > 1) {
 384                 sort(fragtree->splits, nsplits, sizeof(fragtree->splits[0]),
 385                      frag_tree_split_cmp, NULL);
 386         }
 387
 388         dout("fill_fragtree %llx.%llx\n", ceph_vinop(inode));
 389         rb_node = rb_first(&ci->i_fragtree);
 390         for (i = 0; i < nsplits; i++) {
 391                 id = le32_to_cpu(fragtree->splits[i].frag);
 392                 split_by = le32_to_cpu(fragtree->splits[i].by);
 393                 if (split_by == 0 || ceph_frag_bits(id) + split_by > 24) {
 394                         pr_err("fill_fragtree %llx.%llx invalid split %d/%u, "
 395                                "frag %x split by %d\n", ceph_vinop(inode),
 396                                i, nsplits, id, split_by);
 397                         continue;
 398                 }
 399                 frag = NULL;
 400                 while (rb_node) {
 401                         frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 402                         if (ceph_frag_compare(frag->frag, id) >= 0) {
 403                                 if (frag->frag != id)
 404                                         frag = NULL;
 405                                 else
 406                                         rb_node = rb_next(rb_node);
 407                                 break;
 408                         }
 409                         rb_node = rb_next(rb_node);
 410                         /* delete stale split/leaf node */
 411                         if (frag->split_by > 0 ||
 412                             !is_frag_child(frag->frag, prev_frag)) {
 413                                 rb_erase(&frag->node, &ci->i_fragtree);
 414                                 if (frag->split_by > 0)
 415                                         ci->i_fragtree_nsplits--;
 416                                 kfree(frag);
 417                         }
 418                         frag = NULL;
 419                 }
 420                 if (!frag) {
 421                         frag = __get_or_create_frag(ci, id);
 422                         if (IS_ERR(frag))
 423                                 continue;
 424                 }
 425                 if (frag->split_by == 0)
 426                         ci->i_fragtree_nsplits++;
 427                 frag->split_by = split_by;
 428                 dout(" frag %x split by %d\n", frag->frag, frag->split_by);
 429                 prev_frag = frag;
 430         }
 431         while (rb_node) {
 432                 frag = rb_entry(rb_node, struct ceph_inode_frag, node);
 433                 rb_node = rb_next(rb_node);
 434                 /* delete stale split/leaf node */
 435                 if (frag->split_by > 0 ||
 436                     !is_frag_child(frag->frag, prev_frag)) {
 437                         rb_erase(&frag->node, &ci->i_fragtree);
 438                         if (frag->split_by > 0)
 439                                 ci->i_fragtree_nsplits--;
 440                         kfree(frag);
 441                 }
 442         }
 443 out_unlock:
 444         mutex_unlock(&ci->i_fragtree_mutex);
 445         return 0;
 446 }
 447
 448 /*
 449  * initialize a newly allocated inode.
 450  */
 451 struct inode *ceph_alloc_inode(struct super_block *sb)
 452 {
 453         struct ceph_inode_info *ci;
 454         int i;
 455
 456         ci = alloc_inode_sb(sb, ceph_inode_cachep, GFP_NOFS);
 457         if (!ci)
 458                 return NULL;
 459
 460         dout("alloc_inode %p\n", &ci->netfs.inode);
 461
 462         /* Set parameters for the netfs library */
 463         netfs_inode_init(&ci->netfs, &ceph_netfs_ops);
 464
 465         spin_lock_init(&ci->i_ceph_lock);
 466
 467         ci->i_version = 0;
 468         ci->i_inline_version = 0;
 469         ci->i_time_warp_seq = 0;
 470         ci->i_ceph_flags = 0;
 471         atomic64_set(&ci->i_ordered_count, 1);
 472         atomic64_set(&ci->i_release_count, 1);
 473         atomic64_set(&ci->i_complete_seq[0], 0);
 474         atomic64_set(&ci->i_complete_seq[1], 0);
 475         ci->i_symlink = NULL;
 476
 477         ci->i_max_bytes = 0;
 478         ci->i_max_files = 0;
 479
 480         memset(&ci->i_dir_layout, 0, sizeof(ci->i_dir_layout));
 481         memset(&ci->i_cached_layout, 0, sizeof(ci->i_cached_layout));
 482         RCU_INIT_POINTER(ci->i_layout.pool_ns, NULL);
 483
 484         ci->i_fragtree = RB_ROOT;
 485         mutex_init(&ci->i_fragtree_mutex);
 486
 487         ci->i_xattrs.blob = NULL;
 488         ci->i_xattrs.prealloc_blob = NULL;
 489         ci->i_xattrs.dirty = false;
 490         ci->i_xattrs.index = RB_ROOT;
 491         ci->i_xattrs.count = 0;
 492         ci->i_xattrs.names_size = 0;
 493         ci->i_xattrs.vals_size = 0;
 494         ci->i_xattrs.version = 0;
 495         ci->i_xattrs.index_version = 0;
 496
 497         ci->i_caps = RB_ROOT;
 498         ci->i_auth_cap = NULL;
 499         ci->i_dirty_caps = 0;
 500         ci->i_flushing_caps = 0;
 501         INIT_LIST_HEAD(&ci->i_dirty_item);
 502         INIT_LIST_HEAD(&ci->i_flushing_item);
 503         ci->i_prealloc_cap_flush = NULL;
 504         INIT_LIST_HEAD(&ci->i_cap_flush_list);
 505         init_waitqueue_head(&ci->i_cap_wq);
 506         ci->i_hold_caps_max = 0;
 507         INIT_LIST_HEAD(&ci->i_cap_delay_list);
 508         INIT_LIST_HEAD(&ci->i_cap_snaps);
 509         ci->i_head_snapc = NULL;
 510         ci->i_snap_caps = 0;
 511
 512         ci->i_last_rd = ci->i_last_wr = jiffies - 3600 * HZ;
 513         for (i = 0; i < CEPH_FILE_MODE_BITS; i++)
 514                 ci->i_nr_by_mode[i] = 0;
 515
 516         mutex_init(&ci->i_truncate_mutex);
 517         ci->i_truncate_seq = 0;
 518         ci->i_truncate_size = 0;
 519         ci->i_truncate_pending = 0;
 520
 521         ci->i_max_size = 0;
 522         ci->i_reported_size = 0;
 523         ci->i_wanted_max_size = 0;
 524         ci->i_requested_max_size = 0;
 525
 526         ci->i_pin_ref = 0;
 527         ci->i_rd_ref = 0;
 528         ci->i_rdcache_ref = 0;
 529         ci->i_wr_ref = 0;
 530         ci->i_wb_ref = 0;
 531         ci->i_fx_ref = 0;
 532         ci->i_wrbuffer_ref = 0;
 533         ci->i_wrbuffer_ref_head = 0;
 534         atomic_set(&ci->i_filelock_ref, 0);
 535         atomic_set(&ci->i_shared_gen, 1);
 536         ci->i_rdcache_gen = 0;
 537         ci->i_rdcache_revoking = 0;
 538
 539         INIT_LIST_HEAD(&ci->i_unsafe_dirops);
 540         INIT_LIST_HEAD(&ci->i_unsafe_iops);
 541         spin_lock_init(&ci->i_unsafe_lock);
 542
 543         ci->i_snap_realm = NULL;
 544         INIT_LIST_HEAD(&ci->i_snap_realm_item);
 545         INIT_LIST_HEAD(&ci->i_snap_flush_item);
 546
 547         INIT_WORK(&ci->i_work, ceph_inode_work);
 548         ci->i_work_mask = 0;
 549         memset(&ci->i_btime, '\0', sizeof(ci->i_btime));
 550         return &ci->netfs.inode;
 551 }
 552
 553 void ceph_free_inode(struct inode *inode)
 554 {
 555         struct ceph_inode_info *ci = ceph_inode(inode);
 556
 557         kfree(ci->i_symlink);
 558         kmem_cache_free(ceph_inode_cachep, ci);
 559 }
 560
 561 void ceph_evict_inode(struct inode *inode)
 562 {
 563         struct ceph_inode_info *ci = ceph_inode(inode);
 564         struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
 565         struct ceph_inode_frag *frag;
 566         struct rb_node *n;
 567
 568         dout("evict_inode %p ino %llx.%llx\n", inode, ceph_vinop(inode));
 569
 570         percpu_counter_dec(&mdsc->metric.total_inodes);
 571
 572         truncate_inode_pages_final(&inode->i_data);
 573         if (inode->i_state & I_PINNING_FSCACHE_WB)
 574                 ceph_fscache_unuse_cookie(inode, true);
 575         clear_inode(inode);
 576
 577         ceph_fscache_unregister_inode_cookie(ci);
 578
 579         __ceph_remove_caps(ci);
 580
 581         if (__ceph_has_quota(ci, QUOTA_GET_ANY))
 582                 ceph_adjust_quota_realms_count(inode, false);
 583
 584         /*
 585          * we may still have a snap_realm reference if there are stray
 586          * caps in i_snap_caps.
 587          */
 588         if (ci->i_snap_realm) {
 589                 if (ceph_snap(inode) == CEPH_NOSNAP) {
 590                         dout(" dropping residual ref to snap realm %p\n",
 591                              ci->i_snap_realm);
 592                         ceph_change_snap_realm(inode, NULL);
 593                 } else {
 594                         ceph_put_snapid_map(mdsc, ci->i_snapid_map);
 595                         ci->i_snap_realm = NULL;
 596                 }
 597         }
 598
 599         while ((n = rb_first(&ci->i_fragtree)) != NULL) {
 600                 frag = rb_entry(n, struct ceph_inode_frag, node);
 601                 rb_erase(n, &ci->i_fragtree);
 602                 kfree(frag);
 603         }
 604         ci->i_fragtree_nsplits = 0;
 605
 606         __ceph_destroy_xattrs(ci);
 607         if (ci->i_xattrs.blob)
 608                 ceph_buffer_put(ci->i_xattrs.blob);
 609         if (ci->i_xattrs.prealloc_blob)
 610                 ceph_buffer_put(ci->i_xattrs.prealloc_blob);
 611
 612         ceph_put_string(rcu_dereference_raw(ci->i_layout.pool_ns));
 613         ceph_put_string(rcu_dereference_raw(ci->i_cached_layout.pool_ns));
 614 }
 615
 616 static inline blkcnt_t calc_inode_blocks(u64 size)
 617 {
 618         return (size + (1<<9) - 1) >> 9;
 619 }
 620
 621 /*
 622  * Helpers to fill in size, ctime, mtime, and atime.  We have to be
 623  * careful because either the client or MDS may have more up to date
 624  * info, depending on which capabilities are held, and whether
 625  * time_warp_seq or truncate_seq have increased.  (Ordinarily, mtime
 626  * and size are monotonically increasing, except when utimes() or
 627  * truncate() increments the corresponding _seq values.)
 628  */
 629 int ceph_fill_file_size(struct inode *inode, int issued,
 630                         u32 truncate_seq, u64 truncate_size, u64 size)
 631 {
 632         struct ceph_inode_info *ci = ceph_inode(inode);
 633         int queue_trunc = 0;
 634         loff_t isize = i_size_read(inode);
 635
 636         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) > 0 ||
 637             (truncate_seq == ci->i_truncate_seq && size > isize)) {
 638                 dout("size %lld -> %llu\n", isize, size);
 639                 if (size > 0 && S_ISDIR(inode->i_mode)) {
 640                         pr_err("fill_file_size non-zero size for directory\n");
 641                         size = 0;
 642                 }
 643                 i_size_write(inode, size);
 644                 inode->i_blocks = calc_inode_blocks(size);
 645                 /*
 646                  * If we're expanding, then we should be able to just update
 647                  * the existing cookie.
 648                  */
 649                 if (size > isize)
 650                         ceph_fscache_update(inode);
 651                 ci->i_reported_size = size;
 652                 if (truncate_seq != ci->i_truncate_seq) {
 653                         dout("truncate_seq %u -> %u\n",
 654                              ci->i_truncate_seq, truncate_seq);
 655                         ci->i_truncate_seq = truncate_seq;
 656
 657                         /* the MDS should have revoked these caps */
 658                         WARN_ON_ONCE(issued & (CEPH_CAP_FILE_RD |
 659                                                CEPH_CAP_FILE_LAZYIO));
 660                         /*
 661                          * If we hold relevant caps, or in the case where we're
 662                          * not the only client referencing this file and we
 663                          * don't hold those caps, then we need to check whether
 664                          * the file is either opened or mmaped
 665                          */
 666                         if ((issued & (CEPH_CAP_FILE_CACHE|
 667                                        CEPH_CAP_FILE_BUFFER)) ||
 668                             mapping_mapped(inode->i_mapping) ||
 669                             __ceph_is_file_opened(ci)) {
 670                                 ci->i_truncate_pending++;
 671                                 queue_trunc = 1;
 672                         }
 673                 }
 674         }
 675         if (ceph_seq_cmp(truncate_seq, ci->i_truncate_seq) >= 0 &&
 676             ci->i_truncate_size != truncate_size) {
 677                 dout("truncate_size %lld -> %llu\n", ci->i_truncate_size,
 678                      truncate_size);
 679                 ci->i_truncate_size = truncate_size;
 680         }
 681         return queue_trunc;
 682 }
 683
 684 void ceph_fill_file_time(struct inode *inode, int issued,
 685                          u64 time_warp_seq, struct timespec64 *ctime,
 686                          struct timespec64 *mtime, struct timespec64 *atime)
 687 {
 688         struct ceph_inode_info *ci = ceph_inode(inode);
 689         int warn = 0;
 690
 691         if (issued & (CEPH_CAP_FILE_EXCL|
 692                       CEPH_CAP_FILE_WR|
 693                       CEPH_CAP_FILE_BUFFER|
 694                       CEPH_CAP_AUTH_EXCL|
 695                       CEPH_CAP_XATTR_EXCL)) {
 696                 if (ci->i_version == 0 ||
 697                     timespec64_compare(ctime, &inode->i_ctime) > 0) {
 698                         dout("ctime %lld.%09ld -> %lld.%09ld inc w/ cap\n",
 699                              inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
 700                              ctime->tv_sec, ctime->tv_nsec);
 701                         inode->i_ctime = *ctime;
 702                 }
 703                 if (ci->i_version == 0 ||
 704                     ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) > 0) {
 705                         /* the MDS did a utimes() */
 706                         dout("mtime %lld.%09ld -> %lld.%09ld "
 707                              "tw %d -> %d\n",
 708                              inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
 709                              mtime->tv_sec, mtime->tv_nsec,
 710                              ci->i_time_warp_seq, (int)time_warp_seq);
 711
 712                         inode->i_mtime = *mtime;
 713                         inode->i_atime = *atime;
 714                         ci->i_time_warp_seq = time_warp_seq;
 715                 } else if (time_warp_seq == ci->i_time_warp_seq) {
 716                         /* nobody did utimes(); take the max */
 717                         if (timespec64_compare(mtime, &inode->i_mtime) > 0) {
 718                                 dout("mtime %lld.%09ld -> %lld.%09ld inc\n",
 719                                      inode->i_mtime.tv_sec,
 720                                      inode->i_mtime.tv_nsec,
 721                                      mtime->tv_sec, mtime->tv_nsec);
 722                                 inode->i_mtime = *mtime;
 723                         }
 724                         if (timespec64_compare(atime, &inode->i_atime) > 0) {
 725                                 dout("atime %lld.%09ld -> %lld.%09ld inc\n",
 726                                      inode->i_atime.tv_sec,
 727                                      inode->i_atime.tv_nsec,
 728                                      atime->tv_sec, atime->tv_nsec);
 729                                 inode->i_atime = *atime;
 730                         }
 731                 } else if (issued & CEPH_CAP_FILE_EXCL) {
 732                         /* we did a utimes(); ignore mds values */
 733                 } else {
 734                         warn = 1;
 735                 }
 736         } else {
 737                 /* we have no write|excl caps; whatever the MDS says is true */
 738                 if (ceph_seq_cmp(time_warp_seq, ci->i_time_warp_seq) >= 0) {
 739                         inode->i_ctime = *ctime;
 740                         inode->i_mtime = *mtime;
 741                         inode->i_atime = *atime;
 742                         ci->i_time_warp_seq = time_warp_seq;
 743                 } else {
 744                         warn = 1;
 745                 }
 746         }
 747         if (warn) /* time_warp_seq shouldn't go backwards */
 748                 dout("%p mds time_warp_seq %llu < %u\n",
 749                      inode, time_warp_seq, ci->i_time_warp_seq);
 750 }
 751
 752 /*
 753  * Populate an inode based on info from mds.  May be called on new or
 754  * existing inodes.
 755  */
 756 int ceph_fill_inode(struct inode *inode, struct page *locked_page,
 757                     struct ceph_mds_reply_info_in *iinfo,
 758                     struct ceph_mds_reply_dirfrag *dirinfo,
 759                     struct ceph_mds_session *session, int cap_fmode,
 760                     struct ceph_cap_reservation *caps_reservation)
 761 {
 762         struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
 763         struct ceph_mds_reply_inode *info = iinfo->in;
 764         struct ceph_inode_info *ci = ceph_inode(inode);
 765         int issued, new_issued, info_caps;
 766         struct timespec64 mtime, atime, ctime;
 767         struct ceph_buffer *xattr_blob = NULL;
 768         struct ceph_buffer *old_blob = NULL;
 769         struct ceph_string *pool_ns = NULL;
 770         struct ceph_cap *new_cap = NULL;
 771         int err = 0;
 772         bool wake = false;
 773         bool queue_trunc = false;
 774         bool new_version = false;
 775         bool fill_inline = false;
 776         umode_t mode = le32_to_cpu(info->mode);
 777         dev_t rdev = le32_to_cpu(info->rdev);
 778
 779         lockdep_assert_held(&mdsc->snap_rwsem);
 780
 781         dout("%s %p ino %llx.%llx v %llu had %llu\n", __func__,
 782              inode, ceph_vinop(inode), le64_to_cpu(info->version),
 783              ci->i_version);
 784
 785         /* Once I_NEW is cleared, we can't change type or dev numbers */
 786         if (inode->i_state & I_NEW) {
 787                 inode->i_mode = mode;
 788         } else {
 789                 if (inode_wrong_type(inode, mode)) {
 790                         pr_warn_once("inode type changed! (ino %llx.%llx is 0%o, mds says 0%o)\n",
 791                                      ceph_vinop(inode), inode->i_mode, mode);
 792                         return -ESTALE;
 793                 }
 794
 795                 if ((S_ISCHR(mode) || S_ISBLK(mode)) && inode->i_rdev != rdev) {
 796                         pr_warn_once("dev inode rdev changed! (ino %llx.%llx is %u:%u, mds says %u:%u)\n",
 797                                      ceph_vinop(inode), MAJOR(inode->i_rdev),
 798                                      MINOR(inode->i_rdev), MAJOR(rdev),
 799                                      MINOR(rdev));
 800                         return -ESTALE;
 801                 }
 802         }
 803
 804         info_caps = le32_to_cpu(info->cap.caps);
 805
 806         /* prealloc new cap struct */
 807         if (info_caps && ceph_snap(inode) == CEPH_NOSNAP) {
 808                 new_cap = ceph_get_cap(mdsc, caps_reservation);
 809                 if (!new_cap)
 810                         return -ENOMEM;
 811         }
 812
 813         /*
 814          * prealloc xattr data, if it looks like we'll need it.  only
 815          * if len > 4 (meaning there are actually xattrs; the first 4
 816          * bytes are the xattr count).
 817          */
 818         if (iinfo->xattr_len > 4) {
 819                 xattr_blob = ceph_buffer_new(iinfo->xattr_len, GFP_NOFS);
 820                 if (!xattr_blob)
 821                         pr_err("%s ENOMEM xattr blob %d bytes\n", __func__,
 822                                iinfo->xattr_len);
 823         }
 824
 825         if (iinfo->pool_ns_len > 0)
 826                 pool_ns = ceph_find_or_create_string(iinfo->pool_ns_data,
 827                                                      iinfo->pool_ns_len);
 828
 829         if (ceph_snap(inode) != CEPH_NOSNAP && !ci->i_snapid_map)
 830                 ci->i_snapid_map = ceph_get_snapid_map(mdsc, ceph_snap(inode));
 831
 832         spin_lock(&ci->i_ceph_lock);
 833
 834         /*
 835          * provided version will be odd if inode value is projected,
 836          * even if stable.  skip the update if we have newer stable
 837          * info (ours>=theirs, e.g. due to racing mds replies), unless
 838          * we are getting projected (unstable) info (in which case the
 839          * version is odd, and we want ours>theirs).
 840          *   us   them
 841          *   2    2     skip
 842          *   3    2     skip
 843          *   3    3     update
 844          */
 845         if (ci->i_version == 0 ||
 846             ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 847              le64_to_cpu(info->version) > (ci->i_version & ~1)))
 848                 new_version = true;
 849
 850         /* Update change_attribute */
 851         inode_set_max_iversion_raw(inode, iinfo->change_attr);
 852
 853         __ceph_caps_issued(ci, &issued);
 854         issued |= __ceph_caps_dirty(ci);
 855         new_issued = ~issued & info_caps;
 856
 857         /* directories have fl_stripe_unit set to zero */
 858         if (le32_to_cpu(info->layout.fl_stripe_unit))
 859                 inode->i_blkbits =
 860                         fls(le32_to_cpu(info->layout.fl_stripe_unit)) - 1;
 861         else
 862                 inode->i_blkbits = CEPH_BLOCK_SHIFT;
 863
 864         __ceph_update_quota(ci, iinfo->max_bytes, iinfo->max_files);
 865
 866         if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) &&
 867             (issued & CEPH_CAP_AUTH_EXCL) == 0) {
 868                 inode->i_mode = mode;
 869                 inode->i_uid = make_kuid(&init_user_ns, le32_to_cpu(info->uid));
 870                 inode->i_gid = make_kgid(&init_user_ns, le32_to_cpu(info->gid));
 871                 dout("%p mode 0%o uid.gid %d.%d\n", inode, inode->i_mode,
 872                      from_kuid(&init_user_ns, inode->i_uid),
 873                      from_kgid(&init_user_ns, inode->i_gid));
 874                 ceph_decode_timespec64(&ci->i_btime, &iinfo->btime);
 875                 ceph_decode_timespec64(&ci->i_snap_btime, &iinfo->snap_btime);
 876         }
 877
 878         if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) &&
 879             (issued & CEPH_CAP_LINK_EXCL) == 0)
 880                 set_nlink(inode, le32_to_cpu(info->nlink));
 881
 882         if (new_version || (new_issued & CEPH_CAP_ANY_RD)) {
 883                 /* be careful with mtime, atime, size */
 884                 ceph_decode_timespec64(&atime, &info->atime);
 885                 ceph_decode_timespec64(&mtime, &info->mtime);
 886                 ceph_decode_timespec64(&ctime, &info->ctime);
 887                 ceph_fill_file_time(inode, issued,
 888                                 le32_to_cpu(info->time_warp_seq),
 889                                 &ctime, &mtime, &atime);
 890         }
 891
 892         if (new_version || (info_caps & CEPH_CAP_FILE_SHARED)) {
 893                 ci->i_files = le64_to_cpu(info->files);
 894                 ci->i_subdirs = le64_to_cpu(info->subdirs);
 895         }
 896
 897         if (new_version ||
 898             (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) {
 899                 s64 old_pool = ci->i_layout.pool_id;
 900                 struct ceph_string *old_ns;
 901
 902                 ceph_file_layout_from_legacy(&ci->i_layout, &info->layout);
 903                 old_ns = rcu_dereference_protected(ci->i_layout.pool_ns,
 904                                         lockdep_is_held(&ci->i_ceph_lock));
 905                 rcu_assign_pointer(ci->i_layout.pool_ns, pool_ns);
 906
 907                 if (ci->i_layout.pool_id != old_pool || pool_ns != old_ns)
 908                         ci->i_ceph_flags &= ~CEPH_I_POOL_PERM;
 909
 910                 pool_ns = old_ns;
 911
 912                 queue_trunc = ceph_fill_file_size(inode, issued,
 913                                         le32_to_cpu(info->truncate_seq),
 914                                         le64_to_cpu(info->truncate_size),
 915                                         le64_to_cpu(info->size));
 916                 /* only update max_size on auth cap */
 917                 if ((info->cap.flags & CEPH_CAP_FLAG_AUTH) &&
 918                     ci->i_max_size != le64_to_cpu(info->max_size)) {
 919                         dout("max_size %lld -> %llu\n", ci->i_max_size,
 920                                         le64_to_cpu(info->max_size));
 921                         ci->i_max_size = le64_to_cpu(info->max_size);
 922                 }
 923         }
 924
 925         /* layout and rstat are not tracked by capability, update them if
 926          * the inode info is from auth mds */
 927         if (new_version || (info->cap.flags & CEPH_CAP_FLAG_AUTH)) {
 928                 if (S_ISDIR(inode->i_mode)) {
 929                         ci->i_dir_layout = iinfo->dir_layout;
 930                         ci->i_rbytes = le64_to_cpu(info->rbytes);
 931                         ci->i_rfiles = le64_to_cpu(info->rfiles);
 932                         ci->i_rsubdirs = le64_to_cpu(info->rsubdirs);
 933                         ci->i_dir_pin = iinfo->dir_pin;
 934                         ci->i_rsnaps = iinfo->rsnaps;
 935                         ceph_decode_timespec64(&ci->i_rctime, &info->rctime);
 936                 }
 937         }
 938
 939         /* xattrs */
 940         /* note that if i_xattrs.len <= 4, i_xattrs.data will still be NULL. */
 941         if ((ci->i_xattrs.version == 0 || !(issued & CEPH_CAP_XATTR_EXCL))  &&
 942             le64_to_cpu(info->xattr_version) > ci->i_xattrs.version) {
 943                 if (ci->i_xattrs.blob)
 944                         old_blob = ci->i_xattrs.blob;
 945                 ci->i_xattrs.blob = xattr_blob;
 946                 if (xattr_blob)
 947                         memcpy(ci->i_xattrs.blob->vec.iov_base,
 948                                iinfo->xattr_data, iinfo->xattr_len);
 949                 ci->i_xattrs.version = le64_to_cpu(info->xattr_version);
 950                 ceph_forget_all_cached_acls(inode);
 951                 ceph_security_invalidate_secctx(inode);
 952                 xattr_blob = NULL;
 953         }
 954
 955         /* finally update i_version */
 956         if (le64_to_cpu(info->version) > ci->i_version)
 957                 ci->i_version = le64_to_cpu(info->version);
 958
 959         inode->i_mapping->a_ops = &ceph_aops;
 960
 961         switch (inode->i_mode & S_IFMT) {
 962         case S_IFIFO:
 963         case S_IFBLK:
 964         case S_IFCHR:
 965         case S_IFSOCK:
 966                 inode->i_blkbits = PAGE_SHIFT;
 967                 init_special_inode(inode, inode->i_mode, rdev);
 968                 inode->i_op = &ceph_file_iops;
 969                 break;
 970         case S_IFREG:
 971                 inode->i_op = &ceph_file_iops;
 972                 inode->i_fop = &ceph_file_fops;
 973                 break;
 974         case S_IFLNK:
 975                 inode->i_op = &ceph_symlink_iops;
 976                 if (!ci->i_symlink) {
 977                         u32 symlen = iinfo->symlink_len;
 978                         char *sym;
 979
 980                         spin_unlock(&ci->i_ceph_lock);
 981
 982                         if (symlen != i_size_read(inode)) {
 983                                 pr_err("%s %llx.%llx BAD symlink "
 984                                         "size %lld\n", __func__,
 985                                         ceph_vinop(inode),
 986                                         i_size_read(inode));
 987                                 i_size_write(inode, symlen);
 988                                 inode->i_blocks = calc_inode_blocks(symlen);
 989                         }
 990
 991                         err = -ENOMEM;
 992                         sym = kstrndup(iinfo->symlink, symlen, GFP_NOFS);
 993                         if (!sym)
 994                                 goto out;
 995
 996                         spin_lock(&ci->i_ceph_lock);
 997                         if (!ci->i_symlink)
 998                                 ci->i_symlink = sym;
 999                         else
1000                                 kfree(sym); /* lost a race */
1001                 }
1002                 inode->i_link = ci->i_symlink;
1003                 break;
1004         case S_IFDIR:
1005                 inode->i_op = &ceph_dir_iops;
1006                 inode->i_fop = &ceph_dir_fops;
1007                 break;
1008         default:
1009                 pr_err("%s %llx.%llx BAD mode 0%o\n", __func__,
1010                        ceph_vinop(inode), inode->i_mode);
1011         }
1012
1013         /* were we issued a capability? */
1014         if (info_caps) {
1015                 if (ceph_snap(inode) == CEPH_NOSNAP) {
1016                         ceph_add_cap(inode, session,
1017                                      le64_to_cpu(info->cap.cap_id),
1018                                      info_caps,
1019                                      le32_to_cpu(info->cap.wanted),
1020                                      le32_to_cpu(info->cap.seq),
1021                                      le32_to_cpu(info->cap.mseq),
1022                                      le64_to_cpu(info->cap.realm),
1023                                      info->cap.flags, &new_cap);
1024
1025                         /* set dir completion flag? */
1026                         if (S_ISDIR(inode->i_mode) &&
1027                             ci->i_files == 0 && ci->i_subdirs == 0 &&
1028                             (info_caps & CEPH_CAP_FILE_SHARED) &&
1029                             (issued & CEPH_CAP_FILE_EXCL) == 0 &&
1030                             !__ceph_dir_is_complete(ci)) {
1031                                 dout(" marking %p complete (empty)\n", inode);
1032                                 i_size_write(inode, 0);
1033                                 __ceph_dir_set_complete(ci,
1034                                         atomic64_read(&ci->i_release_count),
1035                                         atomic64_read(&ci->i_ordered_count));
1036                         }
1037
1038                         wake = true;
1039                 } else {
1040                         dout(" %p got snap_caps %s\n", inode,
1041                              ceph_cap_string(info_caps));
1042                         ci->i_snap_caps |= info_caps;
1043                 }
1044         }
1045
1046         if (iinfo->inline_version > 0 &&
1047             iinfo->inline_version >= ci->i_inline_version) {
1048                 int cache_caps = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1049                 ci->i_inline_version = iinfo->inline_version;
1050                 if (ceph_has_inline_data(ci) &&
1051                     (locked_page || (info_caps & cache_caps)))
1052                         fill_inline = true;
1053         }
1054
1055         if (cap_fmode >= 0) {
1056                 if (!info_caps)
1057                         pr_warn("mds issued no caps on %llx.%llx\n",
1058                                 ceph_vinop(inode));
1059                 __ceph_touch_fmode(ci, mdsc, cap_fmode);
1060         }
1061
1062         spin_unlock(&ci->i_ceph_lock);
1063
1064         ceph_fscache_register_inode_cookie(inode);
1065
1066         if (fill_inline)
1067                 ceph_fill_inline_data(inode, locked_page,
1068                                       iinfo->inline_data, iinfo->inline_len);
1069
1070         if (wake)
1071                 wake_up_all(&ci->i_cap_wq);
1072
1073         /* queue truncate if we saw i_size decrease */
1074         if (queue_trunc)
1075                 ceph_queue_vmtruncate(inode);
1076
1077         /* populate frag tree */
1078         if (S_ISDIR(inode->i_mode))
1079                 ceph_fill_fragtree(inode, &info->fragtree, dirinfo);
1080
1081         /* update delegation info? */
1082         if (dirinfo)
1083                 ceph_fill_dirfrag(inode, dirinfo);
1084
1085         err = 0;
1086 out:
1087         if (new_cap)
1088                 ceph_put_cap(mdsc, new_cap);
1089         ceph_buffer_put(old_blob);
1090         ceph_buffer_put(xattr_blob);
1091         ceph_put_string(pool_ns);
1092         return err;
1093 }
1094
1095 /*
1096  * caller should hold session s_mutex and dentry->d_lock.
1097  */
1098 static void __update_dentry_lease(struct inode *dir, struct dentry *dentry,
1099                                   struct ceph_mds_reply_lease *lease,
1100                                   struct ceph_mds_session *session,
1101                                   unsigned long from_time,
1102                                   struct ceph_mds_session **old_lease_session)
1103 {
1104         struct ceph_dentry_info *di = ceph_dentry(dentry);
1105         unsigned mask = le16_to_cpu(lease->mask);
1106         long unsigned duration = le32_to_cpu(lease->duration_ms);
1107         long unsigned ttl = from_time + (duration * HZ) / 1000;
1108         long unsigned half_ttl = from_time + (duration * HZ / 2) / 1000;
1109
1110         dout("update_dentry_lease %p duration %lu ms ttl %lu\n",
1111              dentry, duration, ttl);
1112
1113         /* only track leases on regular dentries */
1114         if (ceph_snap(dir) != CEPH_NOSNAP)
1115                 return;
1116
1117         if (mask & CEPH_LEASE_PRIMARY_LINK)
1118                 di->flags |= CEPH_DENTRY_PRIMARY_LINK;
1119         else
1120                 di->flags &= ~CEPH_DENTRY_PRIMARY_LINK;
1121
1122         di->lease_shared_gen = atomic_read(&ceph_inode(dir)->i_shared_gen);
1123         if (!(mask & CEPH_LEASE_VALID)) {
1124                 __ceph_dentry_dir_lease_touch(di);
1125                 return;
1126         }
1127
1128         if (di->lease_gen == atomic_read(&session->s_cap_gen) &&
1129             time_before(ttl, di->time))
1130                 return;  /* we already have a newer lease. */
1131
1132         if (di->lease_session && di->lease_session != session) {
1133                 *old_lease_session = di->lease_session;
1134                 di->lease_session = NULL;
1135         }
1136
1137         if (!di->lease_session)
1138                 di->lease_session = ceph_get_mds_session(session);
1139         di->lease_gen = atomic_read(&session->s_cap_gen);
1140         di->lease_seq = le32_to_cpu(lease->seq);
1141         di->lease_renew_after = half_ttl;
1142         di->lease_renew_from = 0;
1143         di->time = ttl;
1144
1145         __ceph_dentry_lease_touch(di);
1146 }
1147
1148 static inline void update_dentry_lease(struct inode *dir, struct dentry *dentry,
1149                                         struct ceph_mds_reply_lease *lease,
1150                                         struct ceph_mds_session *session,
1151                                         unsigned long from_time)
1152 {
1153         struct ceph_mds_session *old_lease_session = NULL;
1154         spin_lock(&dentry->d_lock);
1155         __update_dentry_lease(dir, dentry, lease, session, from_time,
1156                               &old_lease_session);
1157         spin_unlock(&dentry->d_lock);
1158         ceph_put_mds_session(old_lease_session);
1159 }
1160
1161 /*
1162  * update dentry lease without having parent inode locked
1163  */
1164 static void update_dentry_lease_careful(struct dentry *dentry,
1165                                         struct ceph_mds_reply_lease *lease,
1166                                         struct ceph_mds_session *session,
1167                                         unsigned long from_time,
1168                                         char *dname, u32 dname_len,
1169                                         struct ceph_vino *pdvino,
1170                                         struct ceph_vino *ptvino)
1171
1172 {
1173         struct inode *dir;
1174         struct ceph_mds_session *old_lease_session = NULL;
1175
1176         spin_lock(&dentry->d_lock);
1177         /* make sure dentry's name matches target */
1178         if (dentry->d_name.len != dname_len ||
1179             memcmp(dentry->d_name.name, dname, dname_len))
1180                 goto out_unlock;
1181
1182         dir = d_inode(dentry->d_parent);
1183         /* make sure parent matches dvino */
1184         if (!ceph_ino_compare(dir, pdvino))
1185                 goto out_unlock;
1186
1187         /* make sure dentry's inode matches target. NULL ptvino means that
1188          * we expect a negative dentry */
1189         if (ptvino) {
1190                 if (d_really_is_negative(dentry))
1191                         goto out_unlock;
1192                 if (!ceph_ino_compare(d_inode(dentry), ptvino))
1193                         goto out_unlock;
1194         } else {
1195                 if (d_really_is_positive(dentry))
1196                         goto out_unlock;
1197         }
1198
1199         __update_dentry_lease(dir, dentry, lease, session,
1200                               from_time, &old_lease_session);
1201 out_unlock:
1202         spin_unlock(&dentry->d_lock);
1203         ceph_put_mds_session(old_lease_session);
1204 }
1205
1206 /*
1207  * splice a dentry to an inode.
1208  * caller must hold directory i_rwsem for this to be safe.
1209  */
1210 static int splice_dentry(struct dentry **pdn, struct inode *in)
1211 {
1212         struct dentry *dn = *pdn;
1213         struct dentry *realdn;
1214
1215         BUG_ON(d_inode(dn));
1216
1217         if (S_ISDIR(in->i_mode)) {
1218                 /* If inode is directory, d_splice_alias() below will remove
1219                  * 'realdn' from its origin parent. We need to ensure that
1220                  * origin parent's readdir cache will not reference 'realdn'
1221                  */
1222                 realdn = d_find_any_alias(in);
1223                 if (realdn) {
1224                         struct ceph_dentry_info *di = ceph_dentry(realdn);
1225                         spin_lock(&realdn->d_lock);
1226
1227                         realdn->d_op->d_prune(realdn);
1228
1229                         di->time = jiffies;
1230                         di->lease_shared_gen = 0;
1231                         di->offset = 0;
1232
1233                         spin_unlock(&realdn->d_lock);
1234                         dput(realdn);
1235                 }
1236         }
1237
1238         /* dn must be unhashed */
1239         if (!d_unhashed(dn))
1240                 d_drop(dn);
1241         realdn = d_splice_alias(in, dn);
1242         if (IS_ERR(realdn)) {
1243                 pr_err("splice_dentry error %ld %p inode %p ino %llx.%llx\n",
1244                        PTR_ERR(realdn), dn, in, ceph_vinop(in));
1245                 return PTR_ERR(realdn);
1246         }
1247
1248         if (realdn) {
1249                 dout("dn %p (%d) spliced with %p (%d) "
1250                      "inode %p ino %llx.%llx\n",
1251                      dn, d_count(dn),
1252                      realdn, d_count(realdn),
1253                      d_inode(realdn), ceph_vinop(d_inode(realdn)));
1254                 dput(dn);
1255                 *pdn = realdn;
1256         } else {
1257                 BUG_ON(!ceph_dentry(dn));
1258                 dout("dn %p attached to %p ino %llx.%llx\n",
1259                      dn, d_inode(dn), ceph_vinop(d_inode(dn)));
1260         }
1261         return 0;
1262 }
1263
1264 /*
1265  * Incorporate results into the local cache.  This is either just
1266  * one inode, or a directory, dentry, and possibly linked-to inode (e.g.,
1267  * after a lookup).
1268  *
1269  * A reply may contain
1270  *         a directory inode along with a dentry.
1271  *  and/or a target inode
1272  *
1273  * Called with snap_rwsem (read).
1274  */
1275 int ceph_fill_trace(struct super_block *sb, struct ceph_mds_request *req)
1276 {
1277         struct ceph_mds_session *session = req->r_session;
1278         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1279         struct inode *in = NULL;
1280         struct ceph_vino tvino, dvino;
1281         struct ceph_fs_client *fsc = ceph_sb_to_client(sb);
1282         int err = 0;
1283
1284         dout("fill_trace %p is_dentry %d is_target %d\n", req,
1285              rinfo->head->is_dentry, rinfo->head->is_target);
1286
1287         if (!rinfo->head->is_target && !rinfo->head->is_dentry) {
1288                 dout("fill_trace reply is empty!\n");
1289                 if (rinfo->head->result == 0 && req->r_parent)
1290                         ceph_invalidate_dir_request(req);
1291                 return 0;
1292         }
1293
1294         if (rinfo->head->is_dentry) {
1295                 struct inode *dir = req->r_parent;
1296
1297                 if (dir) {
1298                         err = ceph_fill_inode(dir, NULL, &rinfo->diri,
1299                                               rinfo->dirfrag, session, -1,
1300                                               &req->r_caps_reservation);
1301                         if (err < 0)
1302                                 goto done;
1303                 } else {
1304                         WARN_ON_ONCE(1);
1305                 }
1306
1307                 if (dir && req->r_op == CEPH_MDS_OP_LOOKUPNAME &&
1308                     test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1309                     !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1310                         struct qstr dname;
1311                         struct dentry *dn, *parent;
1312
1313                         BUG_ON(!rinfo->head->is_target);
1314                         BUG_ON(req->r_dentry);
1315
1316                         parent = d_find_any_alias(dir);
1317                         BUG_ON(!parent);
1318
1319                         dname.name = rinfo->dname;
1320                         dname.len = rinfo->dname_len;
1321                         dname.hash = full_name_hash(parent, dname.name, dname.len);
1322                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1323                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1324 retry_lookup:
1325                         dn = d_lookup(parent, &dname);
1326                         dout("d_lookup on parent=%p name=%.*s got %p\n",
1327                              parent, dname.len, dname.name, dn);
1328
1329                         if (!dn) {
1330                                 dn = d_alloc(parent, &dname);
1331                                 dout("d_alloc %p '%.*s' = %p\n", parent,
1332                                      dname.len, dname.name, dn);
1333                                 if (!dn) {
1334                                         dput(parent);
1335                                         err = -ENOMEM;
1336                                         goto done;
1337                                 }
1338                                 err = 0;
1339                         } else if (d_really_is_positive(dn) &&
1340                                    (ceph_ino(d_inode(dn)) != tvino.ino ||
1341                                     ceph_snap(d_inode(dn)) != tvino.snap)) {
1342                                 dout(" dn %p points to wrong inode %p\n",
1343                                      dn, d_inode(dn));
1344                                 ceph_dir_clear_ordered(dir);
1345                                 d_delete(dn);
1346                                 dput(dn);
1347                                 goto retry_lookup;
1348                         }
1349
1350                         req->r_dentry = dn;
1351                         dput(parent);
1352                 }
1353         }
1354
1355         if (rinfo->head->is_target) {
1356                 /* Should be filled in by handle_reply */
1357                 BUG_ON(!req->r_target_inode);
1358
1359                 in = req->r_target_inode;
1360                 err = ceph_fill_inode(in, req->r_locked_page, &rinfo->targeti,
1361                                 NULL, session,
1362                                 (!test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1363                                  !test_bit(CEPH_MDS_R_ASYNC, &req->r_req_flags) &&
1364                                  rinfo->head->result == 0) ?  req->r_fmode : -1,
1365                                 &req->r_caps_reservation);
1366                 if (err < 0) {
1367                         pr_err("ceph_fill_inode badness %p %llx.%llx\n",
1368                                 in, ceph_vinop(in));
1369                         req->r_target_inode = NULL;
1370                         if (in->i_state & I_NEW)
1371                                 discard_new_inode(in);
1372                         else
1373                                 iput(in);
1374                         goto done;
1375                 }
1376                 if (in->i_state & I_NEW)
1377                         unlock_new_inode(in);
1378         }
1379
1380         /*
1381          * ignore null lease/binding on snapdir ENOENT, or else we
1382          * will have trouble splicing in the virtual snapdir later
1383          */
1384         if (rinfo->head->is_dentry &&
1385             !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags) &&
1386             test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1387             (rinfo->head->is_target || strncmp(req->r_dentry->d_name.name,
1388                                                fsc->mount_options->snapdir_name,
1389                                                req->r_dentry->d_name.len))) {
1390                 /*
1391                  * lookup link rename   : null -> possibly existing inode
1392                  * mknod symlink mkdir  : null -> new inode
1393                  * unlink               : linked -> null
1394                  */
1395                 struct inode *dir = req->r_parent;
1396                 struct dentry *dn = req->r_dentry;
1397                 bool have_dir_cap, have_lease;
1398
1399                 BUG_ON(!dn);
1400                 BUG_ON(!dir);
1401                 BUG_ON(d_inode(dn->d_parent) != dir);
1402
1403                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1404                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1405
1406                 BUG_ON(ceph_ino(dir) != dvino.ino);
1407                 BUG_ON(ceph_snap(dir) != dvino.snap);
1408
1409                 /* do we have a lease on the whole dir? */
1410                 have_dir_cap =
1411                         (le32_to_cpu(rinfo->diri.in->cap.caps) &
1412                          CEPH_CAP_FILE_SHARED);
1413
1414                 /* do we have a dn lease? */
1415                 have_lease = have_dir_cap ||
1416                         le32_to_cpu(rinfo->dlease->duration_ms);
1417                 if (!have_lease)
1418                         dout("fill_trace  no dentry lease or dir cap\n");
1419
1420                 /* rename? */
1421                 if (req->r_old_dentry && req->r_op == CEPH_MDS_OP_RENAME) {
1422                         struct inode *olddir = req->r_old_dentry_dir;
1423                         BUG_ON(!olddir);
1424
1425                         dout(" src %p '%pd' dst %p '%pd'\n",
1426                              req->r_old_dentry,
1427                              req->r_old_dentry,
1428                              dn, dn);
1429                         dout("fill_trace doing d_move %p -> %p\n",
1430                              req->r_old_dentry, dn);
1431
1432                         /* d_move screws up sibling dentries' offsets */
1433                         ceph_dir_clear_ordered(dir);
1434                         ceph_dir_clear_ordered(olddir);
1435
1436                         d_move(req->r_old_dentry, dn);
1437                         dout(" src %p '%pd' dst %p '%pd'\n",
1438                              req->r_old_dentry,
1439                              req->r_old_dentry,
1440                              dn, dn);
1441
1442                         /* ensure target dentry is invalidated, despite
1443                            rehashing bug in vfs_rename_dir */
1444                         ceph_invalidate_dentry_lease(dn);
1445
1446                         dout("dn %p gets new offset %lld\n", req->r_old_dentry,
1447                              ceph_dentry(req->r_old_dentry)->offset);
1448
1449                         /* swap r_dentry and r_old_dentry in case that
1450                          * splice_dentry() gets called later. This is safe
1451                          * because no other place will use them */
1452                         req->r_dentry = req->r_old_dentry;
1453                         req->r_old_dentry = dn;
1454                         dn = req->r_dentry;
1455                 }
1456
1457                 /* null dentry? */
1458                 if (!rinfo->head->is_target) {
1459                         dout("fill_trace null dentry\n");
1460                         if (d_really_is_positive(dn)) {
1461                                 dout("d_delete %p\n", dn);
1462                                 ceph_dir_clear_ordered(dir);
1463                                 d_delete(dn);
1464                         } else if (have_lease) {
1465                                 if (d_unhashed(dn))
1466                                         d_add(dn, NULL);
1467                         }
1468
1469                         if (!d_unhashed(dn) && have_lease)
1470                                 update_dentry_lease(dir, dn,
1471                                                     rinfo->dlease, session,
1472                                                     req->r_request_started);
1473                         goto done;
1474                 }
1475
1476                 /* attach proper inode */
1477                 if (d_really_is_negative(dn)) {
1478                         ceph_dir_clear_ordered(dir);
1479                         ihold(in);
1480                         err = splice_dentry(&req->r_dentry, in);
1481                         if (err < 0)
1482                                 goto done;
1483                         dn = req->r_dentry;  /* may have spliced */
1484                 } else if (d_really_is_positive(dn) && d_inode(dn) != in) {
1485                         dout(" %p links to %p %llx.%llx, not %llx.%llx\n",
1486                              dn, d_inode(dn), ceph_vinop(d_inode(dn)),
1487                              ceph_vinop(in));
1488                         d_invalidate(dn);
1489                         have_lease = false;
1490                 }
1491
1492                 if (have_lease) {
1493                         update_dentry_lease(dir, dn,
1494                                             rinfo->dlease, session,
1495                                             req->r_request_started);
1496                 }
1497                 dout(" final dn %p\n", dn);
1498         } else if ((req->r_op == CEPH_MDS_OP_LOOKUPSNAP ||
1499                     req->r_op == CEPH_MDS_OP_MKSNAP) &&
1500                    test_bit(CEPH_MDS_R_PARENT_LOCKED, &req->r_req_flags) &&
1501                    !test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags)) {
1502                 struct inode *dir = req->r_parent;
1503
1504                 /* fill out a snapdir LOOKUPSNAP dentry */
1505                 BUG_ON(!dir);
1506                 BUG_ON(ceph_snap(dir) != CEPH_SNAPDIR);
1507                 BUG_ON(!req->r_dentry);
1508                 dout(" linking snapped dir %p to dn %p\n", in, req->r_dentry);
1509                 ceph_dir_clear_ordered(dir);
1510                 ihold(in);
1511                 err = splice_dentry(&req->r_dentry, in);
1512                 if (err < 0)
1513                         goto done;
1514         } else if (rinfo->head->is_dentry && req->r_dentry) {
1515                 /* parent inode is not locked, be carefull */
1516                 struct ceph_vino *ptvino = NULL;
1517                 dvino.ino = le64_to_cpu(rinfo->diri.in->ino);
1518                 dvino.snap = le64_to_cpu(rinfo->diri.in->snapid);
1519                 if (rinfo->head->is_target) {
1520                         tvino.ino = le64_to_cpu(rinfo->targeti.in->ino);
1521                         tvino.snap = le64_to_cpu(rinfo->targeti.in->snapid);
1522                         ptvino = &tvino;
1523                 }
1524                 update_dentry_lease_careful(req->r_dentry, rinfo->dlease,
1525                                             session, req->r_request_started,
1526                                             rinfo->dname, rinfo->dname_len,
1527                                             &dvino, ptvino);
1528         }
1529 done:
1530         dout("fill_trace done err=%d\n", err);
1531         return err;
1532 }
1533
1534 /*
1535  * Prepopulate our cache with readdir results, leases, etc.
1536  */
1537 static int readdir_prepopulate_inodes_only(struct ceph_mds_request *req,
1538                                            struct ceph_mds_session *session)
1539 {
1540         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1541         int i, err = 0;
1542
1543         for (i = 0; i < rinfo->dir_nr; i++) {
1544                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1545                 struct ceph_vino vino;
1546                 struct inode *in;
1547                 int rc;
1548
1549                 vino.ino = le64_to_cpu(rde->inode.in->ino);
1550                 vino.snap = le64_to_cpu(rde->inode.in->snapid);
1551
1552                 in = ceph_get_inode(req->r_dentry->d_sb, vino);
1553                 if (IS_ERR(in)) {
1554                         err = PTR_ERR(in);
1555                         dout("new_inode badness got %d\n", err);
1556                         continue;
1557                 }
1558                 rc = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1559                                      -1, &req->r_caps_reservation);
1560                 if (rc < 0) {
1561                         pr_err("ceph_fill_inode badness on %p got %d\n",
1562                                in, rc);
1563                         err = rc;
1564                         if (in->i_state & I_NEW) {
1565                                 ihold(in);
1566                                 discard_new_inode(in);
1567                         }
1568                 } else if (in->i_state & I_NEW) {
1569                         unlock_new_inode(in);
1570                 }
1571
1572                 iput(in);
1573         }
1574
1575         return err;
1576 }
1577
1578 void ceph_readdir_cache_release(struct ceph_readdir_cache_control *ctl)
1579 {
1580         if (ctl->page) {
1581                 kunmap(ctl->page);
1582                 put_page(ctl->page);
1583                 ctl->page = NULL;
1584         }
1585 }
1586
1587 static int fill_readdir_cache(struct inode *dir, struct dentry *dn,
1588                               struct ceph_readdir_cache_control *ctl,
1589                               struct ceph_mds_request *req)
1590 {
1591         struct ceph_inode_info *ci = ceph_inode(dir);
1592         unsigned nsize = PAGE_SIZE / sizeof(struct dentry*);
1593         unsigned idx = ctl->index % nsize;
1594         pgoff_t pgoff = ctl->index / nsize;
1595
1596         if (!ctl->page || pgoff != page_index(ctl->page)) {
1597                 ceph_readdir_cache_release(ctl);
1598                 if (idx == 0)
1599                         ctl->page = grab_cache_page(&dir->i_data, pgoff);
1600                 else
1601                         ctl->page = find_lock_page(&dir->i_data, pgoff);
1602                 if (!ctl->page) {
1603                         ctl->index = -1;
1604                         return idx == 0 ? -ENOMEM : 0;
1605                 }
1606                 /* reading/filling the cache are serialized by
1607                  * i_rwsem, no need to use page lock */
1608                 unlock_page(ctl->page);
1609                 ctl->dentries = kmap(ctl->page);
1610                 if (idx == 0)
1611                         memset(ctl->dentries, 0, PAGE_SIZE);
1612         }
1613
1614         if (req->r_dir_release_cnt == atomic64_read(&ci->i_release_count) &&
1615             req->r_dir_ordered_cnt == atomic64_read(&ci->i_ordered_count)) {
1616                 dout("readdir cache dn %p idx %d\n", dn, ctl->index);
1617                 ctl->dentries[idx] = dn;
1618                 ctl->index++;
1619         } else {
1620                 dout("disable readdir cache\n");
1621                 ctl->index = -1;
1622         }
1623         return 0;
1624 }
1625
1626 int ceph_readdir_prepopulate(struct ceph_mds_request *req,
1627                              struct ceph_mds_session *session)
1628 {
1629         struct dentry *parent = req->r_dentry;
1630         struct ceph_inode_info *ci = ceph_inode(d_inode(parent));
1631         struct ceph_mds_reply_info_parsed *rinfo = &req->r_reply_info;
1632         struct qstr dname;
1633         struct dentry *dn;
1634         struct inode *in;
1635         int err = 0, skipped = 0, ret, i;
1636         u32 frag = le32_to_cpu(req->r_args.readdir.frag);
1637         u32 last_hash = 0;
1638         u32 fpos_offset;
1639         struct ceph_readdir_cache_control cache_ctl = {};
1640
1641         if (test_bit(CEPH_MDS_R_ABORTED, &req->r_req_flags))
1642                 return readdir_prepopulate_inodes_only(req, session);
1643
1644         if (rinfo->hash_order) {
1645                 if (req->r_path2) {
1646                         last_hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1647                                                   req->r_path2,
1648                                                   strlen(req->r_path2));
1649                         last_hash = ceph_frag_value(last_hash);
1650                 } else if (rinfo->offset_hash) {
1651                         /* mds understands offset_hash */
1652                         WARN_ON_ONCE(req->r_readdir_offset != 2);
1653                         last_hash = le32_to_cpu(req->r_args.readdir.offset_hash);
1654                 }
1655         }
1656
1657         if (rinfo->dir_dir &&
1658             le32_to_cpu(rinfo->dir_dir->frag) != frag) {
1659                 dout("readdir_prepopulate got new frag %x -> %x\n",
1660                      frag, le32_to_cpu(rinfo->dir_dir->frag));
1661                 frag = le32_to_cpu(rinfo->dir_dir->frag);
1662                 if (!rinfo->hash_order)
1663                         req->r_readdir_offset = 2;
1664         }
1665
1666         if (le32_to_cpu(rinfo->head->op) == CEPH_MDS_OP_LSSNAP) {
1667                 dout("readdir_prepopulate %d items under SNAPDIR dn %p\n",
1668                      rinfo->dir_nr, parent);
1669         } else {
1670                 dout("readdir_prepopulate %d items under dn %p\n",
1671                      rinfo->dir_nr, parent);
1672                 if (rinfo->dir_dir)
1673                         ceph_fill_dirfrag(d_inode(parent), rinfo->dir_dir);
1674
1675                 if (ceph_frag_is_leftmost(frag) &&
1676                     req->r_readdir_offset == 2 &&
1677                     !(rinfo->hash_order && last_hash)) {
1678                         /* note dir version at start of readdir so we can
1679                          * tell if any dentries get dropped */
1680                         req->r_dir_release_cnt =
1681                                 atomic64_read(&ci->i_release_count);
1682                         req->r_dir_ordered_cnt =
1683                                 atomic64_read(&ci->i_ordered_count);
1684                         req->r_readdir_cache_idx = 0;
1685                 }
1686         }
1687
1688         cache_ctl.index = req->r_readdir_cache_idx;
1689         fpos_offset = req->r_readdir_offset;
1690
1691         /* FIXME: release caps/leases if error occurs */
1692         for (i = 0; i < rinfo->dir_nr; i++) {
1693                 struct ceph_mds_reply_dir_entry *rde = rinfo->dir_entries + i;
1694                 struct ceph_vino tvino;
1695
1696                 dname.name = rde->name;
1697                 dname.len = rde->name_len;
1698                 dname.hash = full_name_hash(parent, dname.name, dname.len);
1699
1700                 tvino.ino = le64_to_cpu(rde->inode.in->ino);
1701                 tvino.snap = le64_to_cpu(rde->inode.in->snapid);
1702
1703                 if (rinfo->hash_order) {
1704                         u32 hash = ceph_str_hash(ci->i_dir_layout.dl_dir_hash,
1705                                                  rde->name, rde->name_len);
1706                         hash = ceph_frag_value(hash);
1707                         if (hash != last_hash)
1708                                 fpos_offset = 2;
1709                         last_hash = hash;
1710                         rde->offset = ceph_make_fpos(hash, fpos_offset++, true);
1711                 } else {
1712                         rde->offset = ceph_make_fpos(frag, fpos_offset++, false);
1713                 }
1714
1715 retry_lookup:
1716                 dn = d_lookup(parent, &dname);
1717                 dout("d_lookup on parent=%p name=%.*s got %p\n",
1718                      parent, dname.len, dname.name, dn);
1719
1720                 if (!dn) {
1721                         dn = d_alloc(parent, &dname);
1722                         dout("d_alloc %p '%.*s' = %p\n", parent,
1723                              dname.len, dname.name, dn);
1724                         if (!dn) {
1725                                 dout("d_alloc badness\n");
1726                                 err = -ENOMEM;
1727                                 goto out;
1728                         }
1729                 } else if (d_really_is_positive(dn) &&
1730                            (ceph_ino(d_inode(dn)) != tvino.ino ||
1731                             ceph_snap(d_inode(dn)) != tvino.snap)) {
1732                         struct ceph_dentry_info *di = ceph_dentry(dn);
1733                         dout(" dn %p points to wrong inode %p\n",
1734                              dn, d_inode(dn));
1735
1736                         spin_lock(&dn->d_lock);
1737                         if (di->offset > 0 &&
1738                             di->lease_shared_gen ==
1739                             atomic_read(&ci->i_shared_gen)) {
1740                                 __ceph_dir_clear_ordered(ci);
1741                                 di->offset = 0;
1742                         }
1743                         spin_unlock(&dn->d_lock);
1744
1745                         d_delete(dn);
1746                         dput(dn);
1747                         goto retry_lookup;
1748                 }
1749
1750                 /* inode */
1751                 if (d_really_is_positive(dn)) {
1752                         in = d_inode(dn);
1753                 } else {
1754                         in = ceph_get_inode(parent->d_sb, tvino);
1755                         if (IS_ERR(in)) {
1756                                 dout("new_inode badness\n");
1757                                 d_drop(dn);
1758                                 dput(dn);
1759                                 err = PTR_ERR(in);
1760                                 goto out;
1761                         }
1762                 }
1763
1764                 ret = ceph_fill_inode(in, NULL, &rde->inode, NULL, session,
1765                                       -1, &req->r_caps_reservation);
1766                 if (ret < 0) {
1767                         pr_err("ceph_fill_inode badness on %p\n", in);
1768                         if (d_really_is_negative(dn)) {
1769                                 if (in->i_state & I_NEW) {
1770                                         ihold(in);
1771                                         discard_new_inode(in);
1772                                 }
1773                                 iput(in);
1774                         }
1775                         d_drop(dn);
1776                         err = ret;
1777                         goto next_item;
1778                 }
1779                 if (in->i_state & I_NEW)
1780                         unlock_new_inode(in);
1781
1782                 if (d_really_is_negative(dn)) {
1783                         if (ceph_security_xattr_deadlock(in)) {
1784                                 dout(" skip splicing dn %p to inode %p"
1785                                      " (security xattr deadlock)\n", dn, in);
1786                                 iput(in);
1787                                 skipped++;
1788                                 goto next_item;
1789                         }
1790
1791                         err = splice_dentry(&dn, in);
1792                         if (err < 0)
1793                                 goto next_item;
1794                 }
1795
1796                 ceph_dentry(dn)->offset = rde->offset;
1797
1798                 update_dentry_lease(d_inode(parent), dn,
1799                                     rde->lease, req->r_session,
1800                                     req->r_request_started);
1801
1802                 if (err == 0 && skipped == 0 && cache_ctl.index >= 0) {
1803                         ret = fill_readdir_cache(d_inode(parent), dn,
1804                                                  &cache_ctl, req);
1805                         if (ret < 0)
1806                                 err = ret;
1807                 }
1808 next_item:
1809                 dput(dn);
1810         }
1811 out:
1812         if (err == 0 && skipped == 0) {
1813                 set_bit(CEPH_MDS_R_DID_PREPOPULATE, &req->r_req_flags);
1814                 req->r_readdir_cache_idx = cache_ctl.index;
1815         }
1816         ceph_readdir_cache_release(&cache_ctl);
1817         dout("readdir_prepopulate done\n");
1818         return err;
1819 }
1820
1821 bool ceph_inode_set_size(struct inode *inode, loff_t size)
1822 {
1823         struct ceph_inode_info *ci = ceph_inode(inode);
1824         bool ret;
1825
1826         spin_lock(&ci->i_ceph_lock);
1827         dout("set_size %p %llu -> %llu\n", inode, i_size_read(inode), size);
1828         i_size_write(inode, size);
1829         ceph_fscache_update(inode);
1830         inode->i_blocks = calc_inode_blocks(size);
1831
1832         ret = __ceph_should_report_size(ci);
1833
1834         spin_unlock(&ci->i_ceph_lock);
1835
1836         return ret;
1837 }
1838
1839 void ceph_queue_inode_work(struct inode *inode, int work_bit)
1840 {
1841         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1842         struct ceph_inode_info *ci = ceph_inode(inode);
1843         set_bit(work_bit, &ci->i_work_mask);
1844
1845         ihold(inode);
1846         if (queue_work(fsc->inode_wq, &ci->i_work)) {
1847                 dout("queue_inode_work %p, mask=%lx\n", inode, ci->i_work_mask);
1848         } else {
1849                 dout("queue_inode_work %p already queued, mask=%lx\n",
1850                      inode, ci->i_work_mask);
1851                 iput(inode);
1852         }
1853 }
1854
1855 static void ceph_do_invalidate_pages(struct inode *inode)
1856 {
1857         struct ceph_inode_info *ci = ceph_inode(inode);
1858         u32 orig_gen;
1859         int check = 0;
1860
1861         ceph_fscache_invalidate(inode, false);
1862
1863         mutex_lock(&ci->i_truncate_mutex);
1864
1865         if (ceph_inode_is_shutdown(inode)) {
1866                 pr_warn_ratelimited("%s: inode %llx.%llx is shut down\n",
1867                                     __func__, ceph_vinop(inode));
1868                 mapping_set_error(inode->i_mapping, -EIO);
1869                 truncate_pagecache(inode, 0);
1870                 mutex_unlock(&ci->i_truncate_mutex);
1871                 goto out;
1872         }
1873
1874         spin_lock(&ci->i_ceph_lock);
1875         dout("invalidate_pages %p gen %d revoking %d\n", inode,
1876              ci->i_rdcache_gen, ci->i_rdcache_revoking);
1877         if (ci->i_rdcache_revoking != ci->i_rdcache_gen) {
1878                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1879                         check = 1;
1880                 spin_unlock(&ci->i_ceph_lock);
1881                 mutex_unlock(&ci->i_truncate_mutex);
1882                 goto out;
1883         }
1884         orig_gen = ci->i_rdcache_gen;
1885         spin_unlock(&ci->i_ceph_lock);
1886
1887         if (invalidate_inode_pages2(inode->i_mapping) < 0) {
1888                 pr_err("invalidate_inode_pages2 %llx.%llx failed\n",
1889                        ceph_vinop(inode));
1890         }
1891
1892         spin_lock(&ci->i_ceph_lock);
1893         if (orig_gen == ci->i_rdcache_gen &&
1894             orig_gen == ci->i_rdcache_revoking) {
1895                 dout("invalidate_pages %p gen %d successful\n", inode,
1896                      ci->i_rdcache_gen);
1897                 ci->i_rdcache_revoking--;
1898                 check = 1;
1899         } else {
1900                 dout("invalidate_pages %p gen %d raced, now %d revoking %d\n",
1901                      inode, orig_gen, ci->i_rdcache_gen,
1902                      ci->i_rdcache_revoking);
1903                 if (__ceph_caps_revoking_other(ci, NULL, CEPH_CAP_FILE_CACHE))
1904                         check = 1;
1905         }
1906         spin_unlock(&ci->i_ceph_lock);
1907         mutex_unlock(&ci->i_truncate_mutex);
1908 out:
1909         if (check)
1910                 ceph_check_caps(ci, 0, NULL);
1911 }
1912
1913 /*
1914  * Make sure any pending truncation is applied before doing anything
1915  * that may depend on it.
1916  */
1917 void __ceph_do_pending_vmtruncate(struct inode *inode)
1918 {
1919         struct ceph_inode_info *ci = ceph_inode(inode);
1920         u64 to;
1921         int wrbuffer_refs, finish = 0;
1922
1923         mutex_lock(&ci->i_truncate_mutex);
1924 retry:
1925         spin_lock(&ci->i_ceph_lock);
1926         if (ci->i_truncate_pending == 0) {
1927                 dout("__do_pending_vmtruncate %p none pending\n", inode);
1928                 spin_unlock(&ci->i_ceph_lock);
1929                 mutex_unlock(&ci->i_truncate_mutex);
1930                 return;
1931         }
1932
1933         /*
1934          * make sure any dirty snapped pages are flushed before we
1935          * possibly truncate them.. so write AND block!
1936          */
1937         if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
1938                 spin_unlock(&ci->i_ceph_lock);
1939                 dout("__do_pending_vmtruncate %p flushing snaps first\n",
1940                      inode);
1941                 filemap_write_and_wait_range(&inode->i_data, 0,
1942                                              inode->i_sb->s_maxbytes);
1943                 goto retry;
1944         }
1945
1946         /* there should be no reader or writer */
1947         WARN_ON_ONCE(ci->i_rd_ref || ci->i_wr_ref);
1948
1949         to = ci->i_truncate_size;
1950         wrbuffer_refs = ci->i_wrbuffer_ref;
1951         dout("__do_pending_vmtruncate %p (%d) to %lld\n", inode,
1952              ci->i_truncate_pending, to);
1953         spin_unlock(&ci->i_ceph_lock);
1954
1955         ceph_fscache_resize(inode, to);
1956         truncate_pagecache(inode, to);
1957
1958         spin_lock(&ci->i_ceph_lock);
1959         if (to == ci->i_truncate_size) {
1960                 ci->i_truncate_pending = 0;
1961                 finish = 1;
1962         }
1963         spin_unlock(&ci->i_ceph_lock);
1964         if (!finish)
1965                 goto retry;
1966
1967         mutex_unlock(&ci->i_truncate_mutex);
1968
1969         if (wrbuffer_refs == 0)
1970                 ceph_check_caps(ci, 0, NULL);
1971
1972         wake_up_all(&ci->i_cap_wq);
1973 }
1974
1975 static void ceph_inode_work(struct work_struct *work)
1976 {
1977         struct ceph_inode_info *ci = container_of(work, struct ceph_inode_info,
1978                                                  i_work);
1979         struct inode *inode = &ci->netfs.inode;
1980
1981         if (test_and_clear_bit(CEPH_I_WORK_WRITEBACK, &ci->i_work_mask)) {
1982                 dout("writeback %p\n", inode);
1983                 filemap_fdatawrite(&inode->i_data);
1984         }
1985         if (test_and_clear_bit(CEPH_I_WORK_INVALIDATE_PAGES, &ci->i_work_mask))
1986                 ceph_do_invalidate_pages(inode);
1987
1988         if (test_and_clear_bit(CEPH_I_WORK_VMTRUNCATE, &ci->i_work_mask))
1989                 __ceph_do_pending_vmtruncate(inode);
1990
1991         if (test_and_clear_bit(CEPH_I_WORK_CHECK_CAPS, &ci->i_work_mask))
1992                 ceph_check_caps(ci, 0, NULL);
1993
1994         if (test_and_clear_bit(CEPH_I_WORK_FLUSH_SNAPS, &ci->i_work_mask))
1995                 ceph_flush_snaps(ci, NULL);
1996
1997         iput(inode);
1998 }
1999
2000 /*
2001  * symlinks
2002  */
2003 static const struct inode_operations ceph_symlink_iops = {
2004         .get_link = simple_get_link,
2005         .setattr = ceph_setattr,
2006         .getattr = ceph_getattr,
2007         .listxattr = ceph_listxattr,
2008 };
2009
2010 int __ceph_setattr(struct inode *inode, struct iattr *attr)
2011 {
2012         struct ceph_inode_info *ci = ceph_inode(inode);
2013         unsigned int ia_valid = attr->ia_valid;
2014         struct ceph_mds_request *req;
2015         struct ceph_mds_client *mdsc = ceph_sb_to_client(inode->i_sb)->mdsc;
2016         struct ceph_cap_flush *prealloc_cf;
2017         int issued;
2018         int release = 0, dirtied = 0;
2019         int mask = 0;
2020         int err = 0;
2021         int inode_dirty_flags = 0;
2022         bool lock_snap_rwsem = false;
2023
2024         prealloc_cf = ceph_alloc_cap_flush();
2025         if (!prealloc_cf)
2026                 return -ENOMEM;
2027
2028         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_SETATTR,
2029                                        USE_AUTH_MDS);
2030         if (IS_ERR(req)) {
2031                 ceph_free_cap_flush(prealloc_cf);
2032                 return PTR_ERR(req);
2033         }
2034
2035         spin_lock(&ci->i_ceph_lock);
2036         issued = __ceph_caps_issued(ci, NULL);
2037
2038         if (!ci->i_head_snapc &&
2039             (issued & (CEPH_CAP_ANY_EXCL | CEPH_CAP_FILE_WR))) {
2040                 lock_snap_rwsem = true;
2041                 if (!down_read_trylock(&mdsc->snap_rwsem)) {
2042                         spin_unlock(&ci->i_ceph_lock);
2043                         down_read(&mdsc->snap_rwsem);
2044                         spin_lock(&ci->i_ceph_lock);
2045                         issued = __ceph_caps_issued(ci, NULL);
2046                 }
2047         }
2048
2049         dout("setattr %p issued %s\n", inode, ceph_cap_string(issued));
2050
2051         if (ia_valid & ATTR_UID) {
2052                 dout("setattr %p uid %d -> %d\n", inode,
2053                      from_kuid(&init_user_ns, inode->i_uid),
2054                      from_kuid(&init_user_ns, attr->ia_uid));
2055                 if (issued & CEPH_CAP_AUTH_EXCL) {
2056                         inode->i_uid = attr->ia_uid;
2057                         dirtied |= CEPH_CAP_AUTH_EXCL;
2058                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2059                            !uid_eq(attr->ia_uid, inode->i_uid)) {
2060                         req->r_args.setattr.uid = cpu_to_le32(
2061                                 from_kuid(&init_user_ns, attr->ia_uid));
2062                         mask |= CEPH_SETATTR_UID;
2063                         release |= CEPH_CAP_AUTH_SHARED;
2064                 }
2065         }
2066         if (ia_valid & ATTR_GID) {
2067                 dout("setattr %p gid %d -> %d\n", inode,
2068                      from_kgid(&init_user_ns, inode->i_gid),
2069                      from_kgid(&init_user_ns, attr->ia_gid));
2070                 if (issued & CEPH_CAP_AUTH_EXCL) {
2071                         inode->i_gid = attr->ia_gid;
2072                         dirtied |= CEPH_CAP_AUTH_EXCL;
2073                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2074                            !gid_eq(attr->ia_gid, inode->i_gid)) {
2075                         req->r_args.setattr.gid = cpu_to_le32(
2076                                 from_kgid(&init_user_ns, attr->ia_gid));
2077                         mask |= CEPH_SETATTR_GID;
2078                         release |= CEPH_CAP_AUTH_SHARED;
2079                 }
2080         }
2081         if (ia_valid & ATTR_MODE) {
2082                 dout("setattr %p mode 0%o -> 0%o\n", inode, inode->i_mode,
2083                      attr->ia_mode);
2084                 if (issued & CEPH_CAP_AUTH_EXCL) {
2085                         inode->i_mode = attr->ia_mode;
2086                         dirtied |= CEPH_CAP_AUTH_EXCL;
2087                 } else if ((issued & CEPH_CAP_AUTH_SHARED) == 0 ||
2088                            attr->ia_mode != inode->i_mode) {
2089                         inode->i_mode = attr->ia_mode;
2090                         req->r_args.setattr.mode = cpu_to_le32(attr->ia_mode);
2091                         mask |= CEPH_SETATTR_MODE;
2092                         release |= CEPH_CAP_AUTH_SHARED;
2093                 }
2094         }
2095
2096         if (ia_valid & ATTR_ATIME) {
2097                 dout("setattr %p atime %lld.%ld -> %lld.%ld\n", inode,
2098                      inode->i_atime.tv_sec, inode->i_atime.tv_nsec,
2099                      attr->ia_atime.tv_sec, attr->ia_atime.tv_nsec);
2100                 if (issued & CEPH_CAP_FILE_EXCL) {
2101                         ci->i_time_warp_seq++;
2102                         inode->i_atime = attr->ia_atime;
2103                         dirtied |= CEPH_CAP_FILE_EXCL;
2104                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2105                            timespec64_compare(&inode->i_atime,
2106                                             &attr->ia_atime) < 0) {
2107                         inode->i_atime = attr->ia_atime;
2108                         dirtied |= CEPH_CAP_FILE_WR;
2109                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2110                            !timespec64_equal(&inode->i_atime, &attr->ia_atime)) {
2111                         ceph_encode_timespec64(&req->r_args.setattr.atime,
2112                                                &attr->ia_atime);
2113                         mask |= CEPH_SETATTR_ATIME;
2114                         release |= CEPH_CAP_FILE_SHARED |
2115                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2116                 }
2117         }
2118         if (ia_valid & ATTR_SIZE) {
2119                 loff_t isize = i_size_read(inode);
2120
2121                 dout("setattr %p size %lld -> %lld\n", inode, isize, attr->ia_size);
2122                 if ((issued & CEPH_CAP_FILE_EXCL) && attr->ia_size >= isize) {
2123                         if (attr->ia_size > isize) {
2124                                 i_size_write(inode, attr->ia_size);
2125                                 inode->i_blocks = calc_inode_blocks(attr->ia_size);
2126                                 ci->i_reported_size = attr->ia_size;
2127                                 dirtied |= CEPH_CAP_FILE_EXCL;
2128                                 ia_valid |= ATTR_MTIME;
2129                         }
2130                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2131                            attr->ia_size != isize) {
2132                         req->r_args.setattr.size = cpu_to_le64(attr->ia_size);
2133                         req->r_args.setattr.old_size = cpu_to_le64(isize);
2134                         mask |= CEPH_SETATTR_SIZE;
2135                         release |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL |
2136                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2137                 }
2138         }
2139         if (ia_valid & ATTR_MTIME) {
2140                 dout("setattr %p mtime %lld.%ld -> %lld.%ld\n", inode,
2141                      inode->i_mtime.tv_sec, inode->i_mtime.tv_nsec,
2142                      attr->ia_mtime.tv_sec, attr->ia_mtime.tv_nsec);
2143                 if (issued & CEPH_CAP_FILE_EXCL) {
2144                         ci->i_time_warp_seq++;
2145                         inode->i_mtime = attr->ia_mtime;
2146                         dirtied |= CEPH_CAP_FILE_EXCL;
2147                 } else if ((issued & CEPH_CAP_FILE_WR) &&
2148                            timespec64_compare(&inode->i_mtime,
2149                                             &attr->ia_mtime) < 0) {
2150                         inode->i_mtime = attr->ia_mtime;
2151                         dirtied |= CEPH_CAP_FILE_WR;
2152                 } else if ((issued & CEPH_CAP_FILE_SHARED) == 0 ||
2153                            !timespec64_equal(&inode->i_mtime, &attr->ia_mtime)) {
2154                         ceph_encode_timespec64(&req->r_args.setattr.mtime,
2155                                                &attr->ia_mtime);
2156                         mask |= CEPH_SETATTR_MTIME;
2157                         release |= CEPH_CAP_FILE_SHARED |
2158                                    CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR;
2159                 }
2160         }
2161
2162         /* these do nothing */
2163         if (ia_valid & ATTR_CTIME) {
2164                 bool only = (ia_valid & (ATTR_SIZE|ATTR_MTIME|ATTR_ATIME|
2165                                          ATTR_MODE|ATTR_UID|ATTR_GID)) == 0;
2166                 dout("setattr %p ctime %lld.%ld -> %lld.%ld (%s)\n", inode,
2167                      inode->i_ctime.tv_sec, inode->i_ctime.tv_nsec,
2168                      attr->ia_ctime.tv_sec, attr->ia_ctime.tv_nsec,
2169                      only ? "ctime only" : "ignored");
2170                 if (only) {
2171                         /*
2172                          * if kernel wants to dirty ctime but nothing else,
2173                          * we need to choose a cap to dirty under, or do
2174                          * a almost-no-op setattr
2175                          */
2176                         if (issued & CEPH_CAP_AUTH_EXCL)
2177                                 dirtied |= CEPH_CAP_AUTH_EXCL;
2178                         else if (issued & CEPH_CAP_FILE_EXCL)
2179                                 dirtied |= CEPH_CAP_FILE_EXCL;
2180                         else if (issued & CEPH_CAP_XATTR_EXCL)
2181                                 dirtied |= CEPH_CAP_XATTR_EXCL;
2182                         else
2183                                 mask |= CEPH_SETATTR_CTIME;
2184                 }
2185         }
2186         if (ia_valid & ATTR_FILE)
2187                 dout("setattr %p ATTR_FILE ... hrm!\n", inode);
2188
2189         if (dirtied) {
2190                 inode_dirty_flags = __ceph_mark_dirty_caps(ci, dirtied,
2191                                                            &prealloc_cf);
2192                 inode->i_ctime = attr->ia_ctime;
2193                 inode_inc_iversion_raw(inode);
2194         }
2195
2196         release &= issued;
2197         spin_unlock(&ci->i_ceph_lock);
2198         if (lock_snap_rwsem)
2199                 up_read(&mdsc->snap_rwsem);
2200
2201         if (inode_dirty_flags)
2202                 __mark_inode_dirty(inode, inode_dirty_flags);
2203
2204         if (mask) {
2205                 req->r_inode = inode;
2206                 ihold(inode);
2207                 req->r_inode_drop = release;
2208                 req->r_args.setattr.mask = cpu_to_le32(mask);
2209                 req->r_num_caps = 1;
2210                 req->r_stamp = attr->ia_ctime;
2211                 err = ceph_mdsc_do_request(mdsc, NULL, req);
2212         }
2213         dout("setattr %p result=%d (%s locally, %d remote)\n", inode, err,
2214              ceph_cap_string(dirtied), mask);
2215
2216         ceph_mdsc_put_request(req);
2217         ceph_free_cap_flush(prealloc_cf);
2218
2219         if (err >= 0 && (mask & CEPH_SETATTR_SIZE))
2220                 __ceph_do_pending_vmtruncate(inode);
2221
2222         return err;
2223 }
2224
2225 /*
2226  * setattr
2227  */
2228 int ceph_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
2229                  struct iattr *attr)
2230 {
2231         struct inode *inode = d_inode(dentry);
2232         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
2233         int err;
2234
2235         if (ceph_snap(inode) != CEPH_NOSNAP)
2236                 return -EROFS;
2237
2238         if (ceph_inode_is_shutdown(inode))
2239                 return -ESTALE;
2240
2241         err = setattr_prepare(&init_user_ns, dentry, attr);
2242         if (err != 0)
2243                 return err;
2244
2245         if ((attr->ia_valid & ATTR_SIZE) &&
2246             attr->ia_size > max(i_size_read(inode), fsc->max_file_size))
2247                 return -EFBIG;
2248
2249         if ((attr->ia_valid & ATTR_SIZE) &&
2250             ceph_quota_is_max_bytes_exceeded(inode, attr->ia_size))
2251                 return -EDQUOT;
2252
2253         err = __ceph_setattr(inode, attr);
2254
2255         if (err >= 0 && (attr->ia_valid & ATTR_MODE))
2256                 err = posix_acl_chmod(&init_user_ns, inode, attr->ia_mode);
2257
2258         return err;
2259 }
2260
2261 int ceph_try_to_choose_auth_mds(struct inode *inode, int mask)
2262 {
2263         int issued = ceph_caps_issued(ceph_inode(inode));
2264
2265         /*
2266          * If any 'x' caps is issued we can just choose the auth MDS
2267          * instead of the random replica MDSes. Because only when the
2268          * Locker is in LOCK_EXEC state will the loner client could
2269          * get the 'x' caps. And if we send the getattr requests to
2270          * any replica MDS it must auth pin and tries to rdlock from
2271          * the auth MDS, and then the auth MDS need to do the Locker
2272          * state transition to LOCK_SYNC. And after that the lock state
2273          * will change back.
2274          *
2275          * This cost much when doing the Locker state transition and
2276          * usually will need to revoke caps from clients.
2277          *
2278          * And for the 'Xs' caps for getxattr we will also choose the
2279          * auth MDS, because the MDS side code is buggy due to setxattr
2280          * won't notify the replica MDSes when the values changed and
2281          * the replica MDS will return the old values. Though we will
2282          * fix it in MDS code, but this still makes sense for old ceph.
2283          */
2284         if (((mask & CEPH_CAP_ANY_SHARED) && (issued & CEPH_CAP_ANY_EXCL))
2285             || (mask & (CEPH_STAT_RSTAT | CEPH_STAT_CAP_XATTR)))
2286                 return USE_AUTH_MDS;
2287         else
2288                 return USE_ANY_MDS;
2289 }
2290
2291 /*
2292  * Verify that we have a lease on the given mask.  If not,
2293  * do a getattr against an mds.
2294  */
2295 int __ceph_do_getattr(struct inode *inode, struct page *locked_page,
2296                       int mask, bool force)
2297 {
2298         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2299         struct ceph_mds_client *mdsc = fsc->mdsc;
2300         struct ceph_mds_request *req;
2301         int mode;
2302         int err;
2303
2304         if (ceph_snap(inode) == CEPH_SNAPDIR) {
2305                 dout("do_getattr inode %p SNAPDIR\n", inode);
2306                 return 0;
2307         }
2308
2309         dout("do_getattr inode %p mask %s mode 0%o\n",
2310              inode, ceph_cap_string(mask), inode->i_mode);
2311         if (!force && ceph_caps_issued_mask_metric(ceph_inode(inode), mask, 1))
2312                         return 0;
2313
2314         mode = ceph_try_to_choose_auth_mds(inode, mask);
2315         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
2316         if (IS_ERR(req))
2317                 return PTR_ERR(req);
2318         req->r_inode = inode;
2319         ihold(inode);
2320         req->r_num_caps = 1;
2321         req->r_args.getattr.mask = cpu_to_le32(mask);
2322         req->r_locked_page = locked_page;
2323         err = ceph_mdsc_do_request(mdsc, NULL, req);
2324         if (locked_page && err == 0) {
2325                 u64 inline_version = req->r_reply_info.targeti.inline_version;
2326                 if (inline_version == 0) {
2327                         /* the reply is supposed to contain inline data */
2328                         err = -EINVAL;
2329                 } else if (inline_version == CEPH_INLINE_NONE ||
2330                            inline_version == 1) {
2331                         err = -ENODATA;
2332                 } else {
2333                         err = req->r_reply_info.targeti.inline_len;
2334                 }
2335         }
2336         ceph_mdsc_put_request(req);
2337         dout("do_getattr result=%d\n", err);
2338         return err;
2339 }
2340
2341 int ceph_do_getvxattr(struct inode *inode, const char *name, void *value,
2342                       size_t size)
2343 {
2344         struct ceph_fs_client *fsc = ceph_sb_to_client(inode->i_sb);
2345         struct ceph_mds_client *mdsc = fsc->mdsc;
2346         struct ceph_mds_request *req;
2347         int mode = USE_AUTH_MDS;
2348         int err;
2349         char *xattr_value;
2350         size_t xattr_value_len;
2351
2352         req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETVXATTR, mode);
2353         if (IS_ERR(req)) {
2354                 err = -ENOMEM;
2355                 goto out;
2356         }
2357
2358         req->r_feature_needed = CEPHFS_FEATURE_OP_GETVXATTR;
2359         req->r_path2 = kstrdup(name, GFP_NOFS);
2360         if (!req->r_path2) {
2361                 err = -ENOMEM;
2362                 goto put;
2363         }
2364
2365         ihold(inode);
2366         req->r_inode = inode;
2367         err = ceph_mdsc_do_request(mdsc, NULL, req);
2368         if (err < 0)
2369                 goto put;
2370
2371         xattr_value = req->r_reply_info.xattr_info.xattr_value;
2372         xattr_value_len = req->r_reply_info.xattr_info.xattr_value_len;
2373
2374         dout("do_getvxattr xattr_value_len:%zu, size:%zu\n", xattr_value_len, size);
2375
2376         err = (int)xattr_value_len;
2377         if (size == 0)
2378                 goto put;
2379
2380         if (xattr_value_len > size) {
2381                 err = -ERANGE;
2382                 goto put;
2383         }
2384
2385         memcpy(value, xattr_value, xattr_value_len);
2386 put:
2387         ceph_mdsc_put_request(req);
2388 out:
2389         dout("do_getvxattr result=%d\n", err);
2390         return err;
2391 }
2392
2393
2394 /*
2395  * Check inode permissions.  We verify we have a valid value for
2396  * the AUTH cap, then call the generic handler.
2397  */
2398 int ceph_permission(struct user_namespace *mnt_userns, struct inode *inode,
2399                     int mask)
2400 {
2401         int err;
2402
2403         if (mask & MAY_NOT_BLOCK)
2404                 return -ECHILD;
2405
2406         err = ceph_do_getattr(inode, CEPH_CAP_AUTH_SHARED, false);
2407
2408         if (!err)
2409                 err = generic_permission(&init_user_ns, inode, mask);
2410         return err;
2411 }
2412
2413 /* Craft a mask of needed caps given a set of requested statx attrs. */
2414 static int statx_to_caps(u32 want, umode_t mode)
2415 {
2416         int mask = 0;
2417
2418         if (want & (STATX_MODE|STATX_UID|STATX_GID|STATX_CTIME|STATX_BTIME))
2419                 mask |= CEPH_CAP_AUTH_SHARED;
2420
2421         if (want & (STATX_NLINK|STATX_CTIME)) {
2422                 /*
2423                  * The link count for directories depends on inode->i_subdirs,
2424                  * and that is only updated when Fs caps are held.
2425                  */
2426                 if (S_ISDIR(mode))
2427                         mask |= CEPH_CAP_FILE_SHARED;
2428                 else
2429                         mask |= CEPH_CAP_LINK_SHARED;
2430         }
2431
2432         if (want & (STATX_ATIME|STATX_MTIME|STATX_CTIME|STATX_SIZE|
2433                     STATX_BLOCKS))
2434                 mask |= CEPH_CAP_FILE_SHARED;
2435
2436         if (want & (STATX_CTIME))
2437                 mask |= CEPH_CAP_XATTR_SHARED;
2438
2439         return mask;
2440 }
2441
2442 /*
2443  * Get all the attributes. If we have sufficient caps for the requested attrs,
2444  * then we can avoid talking to the MDS at all.
2445  */
2446 int ceph_getattr(struct user_namespace *mnt_userns, const struct path *path,
2447                  struct kstat *stat, u32 request_mask, unsigned int flags)
2448 {
2449         struct inode *inode = d_inode(path->dentry);
2450         struct super_block *sb = inode->i_sb;
2451         struct ceph_inode_info *ci = ceph_inode(inode);
2452         u32 valid_mask = STATX_BASIC_STATS;
2453         int err = 0;
2454
2455         if (ceph_inode_is_shutdown(inode))
2456                 return -ESTALE;
2457
2458         /* Skip the getattr altogether if we're asked not to sync */
2459         if ((flags & AT_STATX_SYNC_TYPE) != AT_STATX_DONT_SYNC) {
2460                 err = ceph_do_getattr(inode,
2461                                 statx_to_caps(request_mask, inode->i_mode),
2462                                 flags & AT_STATX_FORCE_SYNC);
2463                 if (err)
2464                         return err;
2465         }
2466
2467         generic_fillattr(&init_user_ns, inode, stat);
2468         stat->ino = ceph_present_inode(inode);
2469
2470         /*
2471          * btime on newly-allocated inodes is 0, so if this is still set to
2472          * that, then assume that it's not valid.
2473          */
2474         if (ci->i_btime.tv_sec || ci->i_btime.tv_nsec) {
2475                 stat->btime = ci->i_btime;
2476                 valid_mask |= STATX_BTIME;
2477         }
2478
2479         if (ceph_snap(inode) == CEPH_NOSNAP)
2480                 stat->dev = sb->s_dev;
2481         else
2482                 stat->dev = ci->i_snapid_map ? ci->i_snapid_map->dev : 0;
2483
2484         if (S_ISDIR(inode->i_mode)) {
2485                 if (ceph_test_mount_opt(ceph_sb_to_client(sb), RBYTES)) {
2486                         stat->size = ci->i_rbytes;
2487                 } else if (ceph_snap(inode) == CEPH_SNAPDIR) {
2488                         struct ceph_inode_info *pci;
2489                         struct ceph_snap_realm *realm;
2490                         struct inode *parent;
2491
2492                         parent = ceph_lookup_inode(sb, ceph_ino(inode));
2493                         if (IS_ERR(parent))
2494                                 return PTR_ERR(parent);
2495
2496                         pci = ceph_inode(parent);
2497                         spin_lock(&pci->i_ceph_lock);
2498                         realm = pci->i_snap_realm;
2499                         if (realm)
2500                                 stat->size = realm->num_snaps;
2501                         else
2502                                 stat->size = 0;
2503                         spin_unlock(&pci->i_ceph_lock);
2504                         iput(parent);
2505                 } else {
2506                         stat->size = ci->i_files + ci->i_subdirs;
2507                 }
2508                 stat->blocks = 0;
2509                 stat->blksize = 65536;
2510                 /*
2511                  * Some applications rely on the number of st_nlink
2512                  * value on directories to be either 0 (if unlinked)
2513                  * or 2 + number of subdirectories.
2514                  */
2515                 if (stat->nlink == 1)
2516                         /* '.' + '..' + subdirs */
2517                         stat->nlink = 1 + 1 + ci->i_subdirs;
2518         }
2519
2520         stat->result_mask = request_mask & valid_mask;
2521         return err;
2522 }
2523
2524 void ceph_inode_shutdown(struct inode *inode)
2525 {
2526         struct ceph_inode_info *ci = ceph_inode(inode);
2527         struct rb_node *p;
2528         int iputs = 0;
2529         bool invalidate = false;
2530
2531         spin_lock(&ci->i_ceph_lock);
2532         ci->i_ceph_flags |= CEPH_I_SHUTDOWN;
2533         p = rb_first(&ci->i_caps);
2534         while (p) {
2535                 struct ceph_cap *cap = rb_entry(p, struct ceph_cap, ci_node);
2536
2537                 p = rb_next(p);
2538                 iputs += ceph_purge_inode_cap(inode, cap, &invalidate);
2539         }
2540         spin_unlock(&ci->i_ceph_lock);
2541
2542         if (invalidate)
2543                 ceph_queue_invalidate(inode);
2544         while (iputs--)
2545                 iput(inode);
2546 }